New AI models from OpenAI, Anthropic, Google, and more, with release dates, official sources, and practical notes.
to
Google DeepMind introduced Gemini 4 Hydrogen, a fast model for everyday coding and agentic workflows.
Salesforce released SOV-Transcribe-1, an open-weight model for non-streaming speech recognition.
Zyphra released ZONOS3-Mini, an open-weight multilingual text-to-speech model with voice cloning and fine-grained vocal controls.
Google DeepMind announced AlphaGenome 2, a DNA-sequence model that predicts how genetic variants affect specific human cell types and states.
Baidu released ERNIE 5.5 Flash Preview, a post-trained 280B-parameter model built on ERNIE 5.5 Flash Base, with 14B active parameters and a 1M-token context.
Google introduced Gemini Decision 1, a multimodal model for fast, closed-set choices through the Gemini Decisions API.
MiniMax released MiniMax-H3-Image, an open-weight image generation and editing model built on the H3 family.
Runway made Praxis-1, its world action model, available with open weights under Apache 2.0, after an earlier partner rollout.
Alibaba Cloud introduced Qwen4-Max-Preview, a multimodal model for text generation, image and video understanding, and reasoning.
Subquadratic released SubQ Snap 1.2, a decision model for structured judgments over text and images.
SpeakLeash and ACC Cyfronet AGH released Bielik-11B-v3.1-Instruct, an 11-billion-parameter model trained for 56 languages with emphasis on Polish and other European languages.
Google DeepMind introduced Gemini Robotics 2.5, a vision-language-action model designed to help robots learn a broader range of skills and operate in less structured environments.
The Institute of Foundation Models released PAN 2, an open world model that generates a video of a possible next scene from a starting scene and a described action.
ElevenLabs introduced Scribe v3, a speech-to-text model for recorded audio, with speaker diarization, multilingual support, and controls for terminology and number formatting.
ElevenLabs introduced Scribe v3 Live for real-time speech transcription, adding speaker diarization and context-aware turn detection.
Anthropic introduced Claude Mythos 5.2, an upgraded model for demanding technical work.
Meta introduced Muse Ember, a model designed for repeated agent tasks such as tool calls, coding subtasks, routing, extraction, and long-document parsing.
ByteDance introduced Seedream 5.5, an image-generation model with improved multi-turn editing, native transparent-background output, and more realistic visuals and text rendering.
317 releases shown
Google DeepMind introduced Gemini 4 Hydrogen, a fast model for everyday coding and agentic workflows. Google says it is available across the Gemini API, AI Studio, Google products, and eligible Antigravity subscriptions.
For buildersTry it on routine code execution, test-running, and verification tasks, then reserve deeper architectural decisions for a model intended for long-horizon reasoning.
Salesforce released SOV-Transcribe-1, an open-weight model for non-streaming speech recognition. It supports multilingual transcription, natural-language formatting instructions, and context biasing for domain terms.
For buildersEvaluate it on representative audio containing names, numbers, and domain vocabulary, and review critical transcripts before feeding them into downstream workflows.
Zyphra released ZONOS3-Mini, an open-weight multilingual text-to-speech model with voice cloning and fine-grained vocal controls. Zyphra says the model supports 43 languages and can run locally in browsers and native environments.
For buildersTest voice cloning and language quality with consented reference audio, and check pronunciation and output controls before using generated speech in a production workflow.
Google DeepMind announced AlphaGenome 2, a DNA-sequence model that predicts how genetic variants affect specific human cell types and states. The company says it is available for non-commercial research through the AlphaGenome API, with commercial research access in private preview.
For buildersUse the predictions to prioritize research hypotheses, not as a substitute for experimental validation or clinical judgment.
Baidu released ERNIE 5.5 Flash Preview, a post-trained 280B-parameter model built on ERNIE 5.5 Flash Base, with 14B active parameters and a 1M-token context. Baidu says the preview targets coding, 3D modeling, knowledge work, and content creation.
For buildersTreat this as a preview and verify outputs on your own coding or 3D workflow before relying on it for production work.
Google introduced Gemini Decision 1, a multimodal model for fast, closed-set choices through the Gemini Decisions API. It scores up to 1,000 yes/no, multiple-choice, or rubric questions over shared text, image, audio, video, or PDF input in a single pass.
For buildersUse it for bounded choices such as routing or policy checks, and route low-confidence or consequential decisions to a human review path.
MiniMax released MiniMax-H3-Image, an open-weight image generation and editing model built on the H3 family. It supports text-to-image, edits using reference images, and generation at resolutions up to 16 megapixels.
For buildersTest it on representative image-editing and high-resolution workflows, paying particular attention to reference-image fidelity and the hardware needed to run the weights.
Runway made Praxis-1, its world action model, available with open weights under Apache 2.0, after an earlier partner rollout. The 27B-parameter model predicts video and actions together; Runway says it can be adapted to different robot hardware.
For buildersStart with a controlled simulation or supervised robot setup, and validate safety and task performance on the target hardware before deployment.
Alibaba Cloud introduced Qwen4-Max-Preview, a multimodal model for text generation, image and video understanding, and reasoning. It supports tool calling and structured outputs and is available in preview through Token Plan only.
For buildersTreat this as a time-limited preview, and validate multimodal and tool-calling behavior before relying on it in an application.
Subquadratic released SubQ Snap 1.2, a decision model for structured judgments over text and images. The update adds stronger long-context decision performance and improved consistency when answer options are reordered, according to the company.
For buildersTest option-order stability on representative decisions and check the model's probabilities against known outcomes before automating high-impact decisions.
SpeakLeash and ACC Cyfronet AGH released Bielik-11B-v3.1-Instruct, an 11-billion-parameter model trained for 56 languages with emphasis on Polish and other European languages. The release describes stronger instruction following, tool use, and reasoning than the Bielik 11B v3 line.
For buildersRun a small evaluation in the languages and tool-calling formats you need, and review outputs for factual errors and unsafe content before deployment.
Google DeepMind introduced Gemini Robotics 2.5, a vision-language-action model designed to help robots learn a broader range of skills and operate in less structured environments. The company says it is rolling out to robotics partners.
For buildersFor teams with partner access, assess it on a small, supervised task in the target robot environment and keep people responsible for safety-critical decisions.
The Institute of Foundation Models released PAN 2, an open world model that generates a video of a possible next scene from a starting scene and a described action. IFM says it combines conceptual reasoning with video generation grounded in physical dynamics.
For buildersTreat generated futures as simulations, not forecasts, and compare them with observed outcomes before using them to guide real-world decisions.
ElevenLabs introduced Scribe v3, a speech-to-text model for recorded audio, with speaker diarization, multilingual support, and controls for terminology and number formatting. ElevenLabs reports fewer word errors than Scribe v2 across 102 languages in its evaluation.
For buildersCompare a representative set of recordings with your current transcription workflow, including multiple speakers and specialist terms, and review the transcript before using it downstream.
ElevenLabs introduced Scribe v3 Live for real-time speech transcription, adding speaker diarization and context-aware turn detection. The company reports lower transcription error rates than Scribe v2 Realtime in its evaluations.
For buildersTest it on live audio with interruptions and speaker changes, then check turn boundaries and speaker labels before using results in an automated workflow.
Anthropic introduced Claude Mythos 5.2, an upgraded model for demanding technical work. Anthropic says it improves coding, reasoning, and vision, and makes it available to eligible organizations through its Life Sciences and Cyber Verification Programs.
For buildersIf your organization qualifies, apply through Anthropic's verification programs and evaluate the model only within your approved research scope.
Meta introduced Muse Ember, a model designed for repeated agent tasks such as tool calls, coding subtasks, routing, extraction, and long-document parsing. Meta says it is available through the Meta Model API, with a 1M-token context window and native text, image, audio, and video understanding.
For buildersTry it on a bounded, repeatable workflow and check both the output quality and total token use against your current model before routing larger workloads.
ByteDance introduced Seedream 5.5, an image-generation model with improved multi-turn editing, native transparent-background output, and more realistic visuals and text rendering. The company says it is rolling out across Jimeng, Doubao, Doubao Work, and the Volcano Ark experience center.
For buildersCompare a short series of edits against your current image workflow, checking whether unchanged regions stay stable and whether transparent output fits your production pipeline.
Baidu released ERNIE-5.5-Flash-Base, a 280B sparse mixture-of-experts language model with 14B active parameters per token. It is available as a pretraining checkpoint for downstream adaptation and supports up to 1M-token inference context.
For buildersTreat it as a starting point for adaptation rather than a ready-to-use assistant, and test your own instruction tuning and evaluation before relying on it.
Google made Gemini Omni 1.2 Flash generally available for video generation and editing. Compared with Omni 1.1 Flash, it supports clips up to 15 seconds, multi-turn conversational edits, and stronger reference following; Google lists 720p generation at $0.07 per second.
For buildersCompare a short generated clip and a multi-turn edit against your current workflow, checking reference consistency and any extra access requirements before scaling up.
OpenAI introduced GPT-6 Cyber, a cybersecurity model for finding and validating vulnerabilities while following researcher-defined scope. OpenAI reports improved performance over GPT-5.6 Cyber and says it has been used to identify vulnerabilities and help develop fixes.
For buildersFor approved security research teams, evaluate it only on explicitly authorized targets and compare its findings with a human-reviewed baseline.
Boson AI introduced Higgs Realtime Duplex, a full-duplex interactive multimodal model that listens, watches, and talks while running background reasoning and tool use. Its announcement describes 160 ms interaction chunks and a 45B mixture-of-experts front model; public model access is not documented yet.
For buildersWhen access opens, evaluate its interruption handling and background task results on a bounded, human-reviewed workflow before relying on it.
Thinking Machines Lab introduced Inkling 2 Small on Nov 5, then reported on Nov 12 that its weights were available for download and model integrations had expanded. The model accepts image inputs and supports variable effort.
For buildersIf you have Tinker access, test it on a bounded coding or visual-feedback task and review the model's output and tool use before relying on it.
Ant Group released Ling-3.2-flash, a native multimodal model that accepts image, text, and video input and supports a context window up to 1 million tokens. Ant says it is available through online chat and compatible APIs.
For buildersTry it on a bounded task involving image or video input, then review its output against an independently checked reference before relying on it.
OpenAI introduced Images 2.6 for image generation and editing in ChatGPT, Codex, and the API. It adds faster generation, improved consistency across edits, and API variants GPT-Image-2.6 Flare and Sunburst.
For buildersCompare Flare and Sunburst on the same prompts and editing tasks, checking speed, layout, and consistency against your current workflow.
Cognition introduced SWE-2 Flash, a coding model available in Devin Desktop and CLI. Cognition says it improves on its Inkling-Small base after an added mid-training stage, while model weights are not publicly shared.
For buildersTry it on a bounded coding task in Devin and compare the output, speed, and review effort with your current model before expanding its role.
Tencent announced Hy4 as its latest-generation large language model, adding visual understanding and computer-use capabilities alongside improvements to instruction following, long-context comprehension, reasoning efficiency, and long-horizon agentic tasks. Tencent says model weights and system cards will be released in the coming days.
For buildersTry Hy4 on a bounded workflow that combines image understanding with tool use, and separately evaluate long-context task completion before adopting it.
Pruna AI released P-H3-Image, an open-weights text-to-image and image-editing model built from MiniMax-H3. The team reworked the prompt upsampler, text encoder, diffusion schedule, transformer backbone, and decoder, and says the model generates 4 MP images about five times faster than MiniMax-H3 on a B200.
For buildersEvaluate it on the same prompts and image-editing references you use today, then check the MiniMax H3 Community License and hardware requirements before deployment.
DiffuSpace released Clover-35B-A3B-Preview, a 35B-parameter mixture-of-experts diffusion language model with about 3B active parameters and a thinking mode. DiffuSpace reports results on coding and agent benchmarks; those are the company's evaluations.
For buildersIf evaluating the preview for agent work, compare end-to-end task success and total latency on your own fixed tasks, not just tokens per second.
OpenAI released GPT-6 Rosalind Research for multistep life sciences research, bringing GPT-6 reasoning capabilities with safeguards designed for biological work. Access is limited to approved Research organizations and users.
For buildersFor approved research teams, assess it on a bounded literature or data-analysis workflow and keep scientific conclusions subject to expert review.
The Biological Computing Co. launched mischief 1, an image-to-video model built on Wan 2.2. The company says its neuron-derived optimization generates video five times faster than the base model; it is available through fal and OpenRouter.
For buildersTry it on a small image-to-video batch at the resolutions and clip lengths supported by the endpoint, then compare quality and total generation cost with your current workflow.
InclusionAI released Ling-3.1-flash, a hybrid reasoning model with 560B total parameters, about 25B active per token, and a model context limit of 1M tokens. It builds on Ling-3.0-flash and targets real-world agent, search, office, and software-engineering tasks.
For buildersFor local deployment, start with the model card's hardware and serving guidance, then benchmark representative long-context tasks before choosing a hosted or self-managed setup.
Pika announced H3 Flex, a post-trained version of MiniMax H3 for iterative video creation. Pika says it supports up to 12 multimodal references, synchronized audio, and open weights for local use.
For buildersFor a short video workflow, compare a small reference-guided generation with your current process, and account separately for local setup and Pika-hosted finishing features.
Google DeepMind first introduced Gemini 4 Carbon in limited access on October 28, then expanded availability on November 10 to paying developers and enterprises through its API and business products, with rollout to Google AI Ultra subscribers. The model targets complex, long-horizon agentic work and advanced cybersecurity tasks.
For buildersFor teams admitted to the rollout, evaluate it on a bounded defensive code-review task and independently verify every suggested change before applying it.
NVIDIA released Nemotron 3.5 Transcribe, an open-weight speech-augmented language model for instruction-guided transcription, translation, speaker attribution, and spoken-language identification across 40 languages. NVIDIA says it handles recordings up to one hour and supports word-level timestamps on request.
For buildersEvaluate it on representative recordings with domain-specific names, code-switching, and overlapping speakers. Compare verbatim and cleaned-up output before choosing a transcription workflow.
Particle launched Particle-1, a model post-trained from DeepSeek V4.1 Flash and initially focused on software engineering. Particle says its training approach reduces token use while keeping task success within the base model’s reported confidence intervals on evaluated tasks.
For buildersCompare it with your current coding model on a small, repeatable software task, tracking both completion quality and token use rather than treating efficiency claims as a quality guarantee.
Rime launched Prism, a speech-to-speech model designed for enterprise voice agents. Its generated text can be inspected before speech, while tool calls can run alongside spoken responses.
For buildersEvaluate it on a narrow voice workflow where you need both responsive conversation and a reviewable text checkpoint before an action.
Google Cloud announced Gemini Robotics ER 2 is now available in public preview on the Gemini platform. The enterprise offering pairs the robotics reasoning model with Google Cloud security, orchestration, and lifecycle-management tools.
For buildersFor a robotics prototype, test one low-level action interface as a model tool in simulation and verify task progress before connecting physical hardware.
Google DeepMind announced Genie 4, a world model for real-time interactive simulation with native audio. Users can steer scenes with natural language, and it is rolling out to a limited group of trusted testers.
For buildersNot broadly usable yet; teams in simulation, game prototyping, or agent training should watch for Gemini API early access.
Liquid AI released LFM3, a new generation of open-weight Liquid Foundation Models that puts text, vision, audio, omni, and decision models on one shared backbone. The first release covers six 2B-class models trained from LFM3-1.7B-Base.
For buildersA reasonable starting point for teams that want a small model to fine-tune locally for one specialized task on device.
Microsoft AI launched MAI-Image-2.7 in preview, an image editing model for restyling, relighting, text and symbol edits, and asset restoration. Microsoft AI says it improves on MAI-Image-2.6 across all 10 use cases and 7 editing actions evaluated by Artificial Analysis.
For buildersWorth comparing against your current image editor on product-placement and text-in-image edits, where integration with the scene matters most.
OpenAI released GPT-6.1 Terra, a fast, efficient model in the GPT-6 family that combines visual reasoning and computer use with clearer writing. OpenAI says its standard API prices are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.
For buildersWorth evaluating for high-volume, form-filling or document-entry agents where GPT-6.1 Sol is more model than the task needs.
Volcengine released Doubao-Seedance-1.0-pro-fast, a video-generation model now available on Volcano Ark. The company says it builds on Seedance 1.0 Pro and generates up to about three times faster, with lower cost and multi-shot generation.
For buildersCompare it with Seedance 1.0 Pro on the same short, multi-shot prompt, measuring generation time and reviewing whether the faster output still fits your quality needs.
Block introduced Dessa, a full-duplex speech model trained from scratch on human conversation, with speech input and output and no text or pretrained-model training. It is a research preview, with early access offered to a small group of AI safety researchers.
For buildersIf you can access the preview, evaluate interruption handling, latency, and emotional nuance in a short spoken task, with human review before relying on responses.
Moonshot introduced Kimi K3.1 Pro, the first model in its K3.1 series, alongside K3.1 Pro Highspeed. The company describes a 3T-class open-weight model with a 1-million-token context window and improvements for coding, 3D creation, financial research, and agentic tasks.
For buildersEvaluate both variants on the same long-context coding or research task, comparing output quality, response time, and review effort on your own inputs.
NVIDIA released Nemotron 3.5 Bolt E4B 4GB, an open small language model for conversational agents. NVIDIA describes multi-turn tool use, configurable reasoning, a 64K context window, and deployment on supported local NVIDIA hardware.
For buildersTry it on a narrow local voice-assistant or tool-calling workflow, then measure response quality, latency, and memory use on your target hardware.
Alibaba DAMO Academy introduced RynnUnify, a 16B embodied model pairing an 8B reasoner with an 8B video generator to support visual planning through interleaved video-language reasoning. The project also describes agentic and world-action extensions.
For buildersFor a robotics or simulation prototype, compare direct reasoning with generated visual evidence on a multi-step task, then check whether the predicted scene changes improve the next action.
Canva introduced Craft, its first image-generation foundation model built for design workflows. It is available in Canva and through a public API, producing images up to 2K resolution with composition planning and editable outputs.
For buildersTry the same poster or product-graphic prompt in Craft and your current image model, then compare text placement, composition, and how easily you can revise one element.
InclusionAI open-sourced Ling-3.0-flash-Sante, a healthcare-focused research model with 124B total parameters and 5.1B active per token. Its model card describes medical research, long-document analysis, tool-assisted workflows, and a 256K context; it is not a medical device and requires qualified human review for high-stakes use.
For buildersUse it only in a research sandbox to summarize medical literature with cited source passages, and have a qualified reviewer verify every claim before any consequential use.
Tencent’s WeChat Vision introduced WeVisOne-VL, a vision-language model that combines object-level tasks such as detection, localization, counting and OCR with pixel-level segmentation and matting. The open model family includes 2B and 4B variants.
For buildersTry the 2B and 4B variants on the same image set to compare structured object localization with fine-grained segmentation and matting outputs.
Recursive released Flint, an open mixture-of-experts language model designed for coding agents. Its model card describes repository exploration, bug fixes, feature implementation, tests, and multi-turn tool use; model weights are available from Recursive’s Hugging Face repository.
For buildersEvaluate it on a small, representative repository task, such as diagnosing a failing test and proposing a fix, then verify every change with your usual tests and review process.
Base Labs released mammoth-1 (27B) and elephant-1 (9B), open-source decision models that return probabilities and structured choices, yes/no answers, or scores instead of free-form text. They are available through Baseten Model APIs.
For buildersCompare them on a small set of your own routing, classification, or rubric-scoring cases, and set a review threshold for uncertain outputs before automating decisions.
Mistral introduced Voxtral 3, a family of speech-to-text and text-to-speech models with a voice-agent SDK for real-time, multi-step conversations. Mistral says it is available through its EU-hosted API and for self-deployment.
For buildersEvaluate it on a short, representative voice workflow that combines transcription, a tool call, and a spoken response, then review accuracy, latency, and handoff behavior on your own samples.
Inception Labs released Mercury 3, a diffusion-based language model with a 1M-token context window. It is available through the Inception API Platform and listed distribution partners.
For buildersCompare Mercury 3 with your current model on a representative time-sensitive task, measuring response quality, latency, and review effort on your own inputs.
MiniMax released M3.1 Flash for agentic tasks, coding, computer use, and cybersecurity. It is available through the MiniMax API, M Plan, and MiniMax Code; model weights are expected later, according to MiniMax.
For buildersTest it on a bounded workflow involving several tool steps or a coding task, then compare task completion, latency, and human review needs with your current model.
Reflection released Beam’s open weights on October 16 under the Apache 2.0 license, following its earlier early-access announcement. The 501B-parameter sparse MoE model has 23B active parameters and is built for coding, reasoning, and agentic workloads; it can be self-hosted or accessed through Reflection and partners.
For buildersCompare the released weights or a hosted endpoint on a bounded coding task from your own workflow, tracking successful completion, tool-use errors, latency, and review effort.
Perplexity released two open-weight multimodal decision models: pplx-decider-v1.2 and pplx-decider-flash-v1.2. They process multiple questions about the same input in one pass and return structured decisions with probabilities, rather than free-form text.
For buildersTry both sizes on a small set of your own routing or classification cases, then review probability calibration and uncertain decisions with a human before automating actions.
OpenAI began rolling out GPT-Live-1.1 and the mini GPT-Live-1.1 Quip model in ChatGPT Voice, with more natural accents and additional language-specific voices. Availability varies by region and account; the announcement describes a ChatGPT Voice rollout, not general API access.
For buildersIf the models are available to your account, compare them on short voice interactions across supported languages, checking transcription, turn-taking, and response quality.
The Institute of Foundation Models and Inception42 released Jais 3, a family of six open Arabic language models from 900 million to 375 billion parameters, alongside Jais Speech for Arabic speech recognition. The models support Arabic across regional dialects and are available on Hugging Face.
For buildersEvaluate the model family on representative Modern Standard Arabic and regional-dialect prompts, plus speech samples if using Jais Speech; check transcription and task completion against a reviewed baseline.
Liquid AI released d1-ultra, a multimodal decision model that accepts text and images and returns probabilities for yes/no, choice, or scoring questions without generating tokens. It is available through the Liquid AI API; performance comparisons in the announcement are provider-reported.
For buildersTry it on a narrow classification or visual-inspection task with labeled examples, then check its decisions and probability calibration against human-reviewed outcomes.
Meta released the open-weight Muse Spark Open, a natively multimodal model with 816B total parameters, 42B active parameters, and a 1M-token context window. Meta says it is built for coding and agentic work and is available through hosting platforms and downloadable weights.
For buildersCompare it with your current model on a bounded coding or multimodal agent task, tracking task completion and review effort on your own examples.
NVIDIA released Nemotron 3.5 Super VL and Bolt 74B-A9B 24GB on October 15, then released Nemotron 3.5 Bolt E4B 4GB on October 23. The family targets multimodal agent workflows and local conversational agents; NVIDIA describes E4B as a small open model with multi-turn tool use and a 64K context window.
For buildersEvaluate the available Super VL or Bolt model on a representative visual-document or local-agent task, and compare accuracy, latency, and hardware needs with your current setup.
xAI released Grok Voice 3.0, a speech-to-speech model available through its Voice Agent API. The API is unchanged, and the grok-voice-latest alias now points to version 3.0.
For buildersCompare it with your current voice agent on representative calls, including interruptions and varied accents, and review task completion, latency, and handoff quality.
Sqwish Labs released the kettei decision-model family: kettei fat and chub handle text and images, while core and dot are text-only; core and dot are open source. kettei fat, chub and core are available through OneHot, with fine-tuning for fat in early preview.
For buildersTry an exact, bounded decision task such as routing or evidence checks with fixed choices, then compare its scores and review workload against the current classifier or LLM workflow.
Google announced broader availability for Gemini 4 Argon, a model designed for long-horizon reasoning across software engineering, enterprise knowledge work, and cybersecurity defense. Starting October 13, developers can use it through the paid Gemini API and Android Studio, with access also available in Google AI Studio and Antigravity for AI Ultra subscribers, and in Gemini app and Search AI Mode for eligible subscribers.
For buildersIf you have access, compare Argon with your current model on a representative long-horizon coding or document-work task, tracking completion quality and review effort rather than relying on provider benchmark claims.
Noiz AI and HKUST introduced Helix World, a real-time interactive world model that generates video and spatial stereo audio together from a starting image, conditioned on camera movement and keyboard input. A preview checkpoint is available on Hugging Face alongside a public demo, paper, and code.
For buildersTry the preview on a small set of starting images and camera movements, then review whether generated motion and stereo sound remain coherent as the viewpoint changes.
Meta introduced Muse Wonder and Muse Wonder Full, multimodal models for personal and professional agentic tasks that use Active Memory and Active Focus. Muse Wonder is available through the Meta Model API and Muse Code; Muse Wonder Full is available to Muse Code subscribers.
For buildersCompare the models on a small set of multi-step workflows that require recalling earlier context, using tools, and recovering from a blocker. Track task completion and review consequential actions before relying on them.
Cloudflare released Clef-omni, an open-weight decision model that accepts audio, video, image, and text inputs in one API call. Its release builds on the existing Clef family and makes its weights available on Hugging Face.
For buildersTry Clef-omni on a workflow that currently chains transcription with image or video analysis, then compare decision quality and latency against that pipeline.
Microsoft introduced Microsoft-Decision-1, a decision-scoring model for structured tasks such as routing, classification, prioritization, and verification. It returns calibrated scores for fixed choices and is available through Microsoft Foundry and OpenRouter.
For buildersEvaluate it on a representative routing or classification task with a fixed set of choices, and measure decision quality and latency against the current approach.
JetBrains released Mellum2.1, an open 12B mixture-of-experts model for coding agents. Its post-training uses reinforcement learning in real coding environments, enabling it to explore repositories, edit files, and check its changes; the model is available on Hugging Face under Apache 2.0.
For buildersTry it on a representative codebase task that includes diagnosing a failing test, proposing a patch, and checking the result; compare quality and latency with your current coding model.
Anthropic released Claude Haiku 5.5 for high-volume, cost-sensitive tasks including summaries, database queries, classification, and coding subagent work. It is available now across the Claude Platform and cloud providers, with an adjustable effort setting.
For buildersCompare Haiku 5.5 with a larger model on representative classification, summarization, or coding-subagent tasks, tracking quality, latency, and cost before routing production work.
Liquid AI released two open-weight decision models, d1-3B and d1-omni-600M, which produce answers in a single forward pass rather than generating tokens. d1-3B supports text and image inputs, while the experimental d1-omni-600M supports text+image or text+audio; both are available on Hugging Face.
For buildersTest the d1 models on a bounded decision task using representative text, image, or audio inputs, then compare answer quality and latency with your current model before adopting them.
Perplexity released two late-interaction embedding models, pplx-embed-v2-late-0.6b and pplx-embed-v2-late-9b, for text and image retrieval. They retain token-level vectors and share an embedding space, and both are publicly available on Hugging Face.
For buildersCompare the 0.6B and 9B variants on a representative set of text queries and scanned or rendered documents; measure retrieval relevance alongside index size and query latency.
Google launched EmbeddingGemma 2, an open multimodal embedding model that maps text, code, images, video, and audio into a shared embedding space. Its 740M-parameter weights are available under Apache 2.0 for on-device search and retrieval.
For buildersCompare local retrieval with your current embedding model using the same text, code, image, audio, and video queries; review relevance, latency, and device memory before adopting it.
Mistral released improved Mistral Large 4 weights on October 29, following its public preview earlier in October. The natively multimodal model has a 1 trillion parameter scale and 52 billion active parameters, according to Mistral.
For buildersTest the released weights on representative coding and defensive-security tasks, then compare output quality and token use against the public preview or another model before deployment.
Google’s Nano Banana 2.1 is an image-generation and editing model in the Gemini 3 family. Its model card lists text and image inputs, image and text outputs, and distribution through Gemini, Google AI Studio, the Gemini API, Search AI Mode, Ads, Flow, and Stitch.
For buildersCompare it with your current image model on a small set of prompts that test text rendering, image edits, and multi-reference consistency; review outputs for fidelity and instruction-following errors.
Perplexity released pplx-decider-v1.1-27b, an updated open-weights multimodal decision model. It is designed to return structured decisions such as yes/no, choice, or score outputs rather than generated text.
For buildersEvaluate it on a small set of representative classification or routing examples, comparing decision probabilities with your current approach and checking uncertain cases with human review.
Amazon announced general availability of Nova 2.5 Sonic, a real-time speech-to-speech model for voice agents. The release adds improved reasoning, instruction following, tool-calling accuracy, and lower latency, according to AWS.
For buildersEvaluate it with representative voice-agent conversations that require turn-taking, multi-step instructions, and tool use, then review accuracy and handoff behavior.
Reka released a research preview of Rho-1, a 19B omni-reasoning model trained from scratch. The company says it can understand and generate text, images, and video, reason across them, and take actions within one model.
For buildersIf you can access the preview, test a task that combines visual understanding, generation, and follow-up editing, then review each output for consistency and errors.
Aleph Alpha released Kolibri, an English-German open-weight Mixture-of-Experts language model with 78B total parameters, 3B active parameters, and a context length of up to 1 million tokens. Full weights are available under Apache 2.0; the provider describes it as specialized for regulated, mission-critical work.
For buildersEvaluate it locally on representative English and German documents, including long-context retrieval and abstention cases, before considering it for regulated workflows.
Ai2 released AstaBrief 8B, an open-weights model that turns a research question and retrieved literature into a cited scientific report in one pass. The model and training data are available for download; it is also offered as Fast mode in Ai2’s Asta platform.
For buildersTry it on a small set of research questions with known source papers, then check citation support, omitted caveats, and whether the report is useful as a first draft.
Cactus released Whistle, a 16.9 MB on-device speech-recognition model for CPU deployment. It transcribes up to 30 seconds of audio in seven languages and also returns word timestamps and speech embeddings without sending audio off-device.
For buildersTest it on short, representative clips on the target device, measuring word errors, timestamps, latency, and performance on names before using transcripts to trigger actions.
Decagon introduced Chord, its first speech model, trained and post-trained for customer conversations and used by its Voice 3 system. The company says Chord shapes speech phrase by phrase and is trained on licensed data and consented voice talent; Voice 3 supports more than 70 languages.
For buildersIf evaluating a customer-service voice stack, test real turn-taking, interruptions, confirmation details, pronunciation, and handoff behavior across representative calls.
Cloudflare released Clef and Clef-flash, open-weight decision models hosted on Workers AI that return probabilities for typed choices instead of generating text. Clef adds image input and a 64K context window; the weights are available under Apache 2.0.
For buildersEvaluate one bounded task such as support-ticket routing against your current classifier or model, measuring decision quality, confidence calibration, latency, and how often a human should review the result.
Tavus introduced Griffin, a real-time video-to-video model that combines perception, conversational behavior, and audiovisual generation. Only Griffin-Lite is currently available, as a research preview for select early testers; a wider release is still to come.
For buildersFor a face-to-face agent prototype, evaluate whether live video and interruption handling improve conversation flow, and assess disclosure and consent requirements before user testing.
Microsoft introduced MAI-Transcribe-2-Streaming for live speech transcription and two speech-generation models, MAI-Voice-2.1 and MAI-Voice-2.1-Flash. Transcribe supports 60 languages with automatic detection and partial transcripts in just over 100 ms; the voice models support multilingual speech generation, with Flash designed for low-latency workloads.
For buildersPrototype a voice-agent loop using representative call audio, then compare transcription latency and short-utterance accuracy alongside response quality and voice consistency across the languages you need.
Cohere released Embed 5, a family of multimodal embedding models in Pro and Fast tiers. Both support text and image inputs, more than 100 languages, and a 128K-token context; their shared embedding space lets teams index with Pro and query with either tier.
For buildersTest both tiers on representative text, image-rich documents, and multilingual queries; compare retrieval quality, latency, and index consistency before choosing production roles.
Perplexity introduced pplx-embed-v2-context-9b-preview, a contextual embedding model that encodes document chunks with document-wide context and returns one embedding per chunk. It is publicly available as a preview; the provider cautions that weights and interface may change and that preview embeddings should not be mixed with later versions.
For buildersTest it on long documents where answer passages depend on distant context, comparing retrieval of both answer-bearing and supporting passages against your current chunk embeddings.
OpenAI released GPT-6.1 Sol, an upgrade to GPT-6 Sol that it says nearly matches GPT-6 Astra on agentic coding, computer use, and professional work at one-fifth of Astra's standard token price. It is available today in ChatGPT Work and Codex and through the API as gpt-6.1-sol.
For buildersUse GPT-6.1 Sol for everyday coding and agent workloads and escalate to GPT-6 Astra only when tasks fail, since Sol costs $2 input and $10 output per million tokens versus Astra's $10 and $50.
NVIDIA released Kumo Tabular, an open foundation model for tabular classification and regression. It predicts labels for new rows from labeled examples and is available in Small, Medium, and Large sizes.
For buildersEvaluate it on a held-out tabular task against a strong tree-based baseline, especially when labeled examples are limited.
AssemblyAI released Universal-3.6 Pro Realtime, an upgrade to Universal-3.5 Pro for streaming speech recognition. It supports 32 languages with automatic detection and is designed to handle short responses, multiple speakers, accents, and noisy telephony audio at the previous version’s latency.
For buildersTest it on a representative set of noisy calls with short confirmations, names, and code-switching, and measure transcription errors before connecting its output to automated actions.
Voltropy announced its first model family, Vast-10M, in Flash, Medium, and Pro variants, using its Voltropy Scalable Attention method to extend transformer context. The company is offering early access; general availability is not confirmed.
For buildersIf admitted to early access, test recall on a large document or code corpus against a smaller-context baseline.
Anthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family, positioned as a faster and lower-cost complement to Opus 5.5 for everyday coding and document work. It is priced the same as Sonnet 5 at $2 per million input tokens and $10 per million output tokens and generates output more than 30% faster.
For buildersTry Sonnet 5.5 as a faster default for well-scoped coding and document tasks, and measure cost per finished task on your own workloads to confirm the claimed token savings.
Hcompany released Holo4, a family of agentic models in two sizes: a 27B dense model and a 35B MoE (35B-A3B) model, plus an updated Holotron4 Nano. They work across GUIs, code, MCP, and APIs, and are available through the H Models API with open weights on Hugging Face.
For buildersEvaluate Holo4 through the H Models API on a multi-step task that mixes screen interaction, code execution, and API tool calls, since it is built to switch between interfaces within one workflow.
Kling AI announced Kling 4.0, its next-generation video generation model, with single clips up to 30 seconds, 4K and 1080p 10-bit HDR output, and up to 15 multimodal reference inputs. Kling 4.0 Flash opened limited early access on September 28, and the full release is planned for October.
For buildersWhen Kling 4.0 reaches general availability, test its multi-keyframe storyboard control for long-form and commercial video work, and compare per-generation cost and clip length with ByteDance Seedance 2.5.
AutoTrust AI introduced JEV-27B, an open-weights model that pairs calibrated, typed one-pass decisions with a Qwen3.8-27B text-generation and reasoning path. The project publishes the weights under Apache-2.0.
For buildersEvaluate it for triage or routing by checking calibration on held-out labeled examples and the quality of the generation path.
Meituan launched LongCat-2.5 Preview on its API platform, adding native image understanding to its long-context model. It keeps a 1.6T parameter MoE architecture with about 48B active parameters, a 1M token context window, and 128K token output, and targets long-horizon agent tasks across browsers, terminals, and desktop apps.
For buildersTest LongCat-2.5 Preview through Meituan's API for long-context agent work such as reading and modifying a large codebase, since its endpoints are compatible with Anthropic and OpenAI style calls.
Perceptron released Mk1.5, a model for controlling embodied agents including drones, robotic dogs, smart glasses, and phones. It adds native audio input, video object tracking with timestamped geometries, and egocentric video understanding, with a 32K token context window.
For buildersTry it through the Perceptron Platform API at $0.15/M input tokens for tasks that require real-time video tracking or embodied agent control without platform-specific retraining.
Stanford and NVIDIA researchers released CLM-8B, an open contrastive language model that selects the best matching action for a state instead of generating tokens, which makes it fast for bounded agent decisions. It is available under Apache 2.0 with open weights and targets computer use, tool calling, and verification tasks.
For buildersUse CLM-8B as a fast action selector or verifier in agent pipelines where the model chooses between a fixed set of tools or candidate outputs, and cache action embeddings to cut latency on repeated decisions.
Black Forest Labs released FLUX 3 Action, an open-weight 7B world action model that predicts future actions from camera frames and text instructions. Black Forest Labs reports that a DROID fine-tune placed first on the RoboLab-120 benchmark at a 42.92% success rate.
For buildersUse FLUX 3 Action through the LeRobot integration to fine-tune a robotics policy for your own robot arm or test it in simulation through NVIDIA Isaac Sim before deploying on hardware.
Google launched Gemini 3.8 Flash TTS and Flash-Lite TTS, expressive text-to-speech models that create custom voices from prompts, replicate a voice from a 30-second sample, and control delivery line by line. Both are available today in Google AI Studio and the Gemini API, with Flash-Lite aimed at high-volume dubbing and localization.
For buildersPrompt a custom voice in Google AI Studio and rehearse line-by-line delivery in the dual-speaker editor before wiring it into your app through the Gemini API.
Meta introduced Muse Realtime Avatar, an embodiment model that turns Muse Realtime Voice into expressive interactive avatars with synchronized lip movement and gestures. It powers live video conversations in the Muse app, which Meta says are rolling out, and generated video is watermarked with Video Seal.
For buildersRun a short live conversation with a reference photo or illustration and check whether the avatar keeps consistent appearance and lip sync across turns before building a branded character around it.
NVIDIA released Nemotron 3 Diarization, a 100-million-parameter open-weight model that identifies who spoke when in real time for up to eight speakers, independent of speech recognition. NVIDIA reports it ranks first in VoiceArena's initial Diarization-Bench with a 14.72 percent diarization error rate.
For buildersDownload the open weights and pair its speaker timestamps with a transcription model to build meeting notes or call summaries without sending audio to a cloud API.
NVIDIA released NV-Reason-CT, an open research vision language model that applies chain of thought reasoning to 3D CT scans. It generates structured reports and supports multi-step follow-up questions for chest and abdominal CT, and is positioned as a foundation model for researchers rather than a cleared clinical product.
For buildersYou can post-train NV-Reason-CT on the open weights for CT-specific applications, but it is a research foundation model, not approved for clinical diagnostic use.
OpenAI released GPT-6 Sol for coding and professional work, with API pricing of $2 per million input tokens and $10 per million output tokens. Access is rolling out in ChatGPT Work and Codex; it is not yet available in Chat.
For buildersEvaluate a coding or research workflow on your own tasks, comparing quality and total cost with the model you currently use.
Anthropic released Claude Opus 5.5, the first model in its Claude 5.5 family, for long-running coding and knowledge work. Company pricing is $4 per million input tokens and $20 per million output tokens, with cache reads at $0.20 per million.
For buildersRun the same long coding or knowledge-work job on Opus 5.5 and Opus 5, then compare total token cost and whether the cheaper cache-read rate actually shows up in the bill.
OpenAI released GPT-6 Luna alongside Sol as a lower-cost model, with API pricing of $0.10 per million input tokens and $0.50 per million output tokens. Free and Go users can access Luna in the desktop app.
For buildersTest a high-volume workflow such as classifying support requests or extracting structured fields from incoming documents.
Tencent released Hy Image 3.5 Preview, an image generation model supporting text-to-image, multi-image reference generation, and multi-turn conversational editing, with output up to 4K through the API. It is available through Tencent Cloud TokenHub from about $0.024 per image.
For buildersIntegrate Hy Image 3.5 Preview's messages-based API for iterative image editing workflows where users refine visuals through conversation rather than one-shot generation, useful for e-commerce and content production pipelines.
Xiaomi open-sourced the MiMo-V2.6-Pro and MiMo-V2.6-Flash multimodal LLMs, with Pro scoring 46 on the AI Composite Intelligence Index to become the top open-weight model globally. Both models use scaled reinforcement learning for recursive self-improvement and are available as open weights.
For buildersTry MiMo-V2.6-Pro or MiMo-V2.6-Flash through their open-weight releases for cost-efficient multimodal tasks like image understanding, 3D perception, and computer use, especially if you want to avoid per-token API costs from closed-source providers.
Alibaba updated its Qwen-Audio 3.1 speech suite at Apsara 2026, covering automatic speech recognition, text-to-speech, and real-time voice interaction models with enhanced voice AI capabilities.
For buildersWire the Realtime model into a voice agent prototype and test turn-taking and interruption handling in a noisy room before scaling it.
Alibaba unveiled Qwen-Audio-3.1-TTS-Next at Apsara 2026, an audio generation model that creates cinematic soundscapes by blending dialogue and ambient sound from a text script. It targets audiobooks, film, TV, podcasts, and games.
For buildersFeed it a short script with dialogue and scene cues and compare the mixed audio against separate speech and effects pipelines for your project.
Alibaba announced Qwen-Image 3.1, an image generation model for design and e-commerce marketing with native transparent-background generation and image editing. It is scheduled to launch later this year.
For buildersPlan a hero-image and product-photo workflow around transparent-background generation once it ships, and compare it against your current image model.
Alibaba debuted Qwen3.8-LiveTranslate at Apsara 2026, a simultaneous interpretation model for real-time translation designed for smoother, more responsive live translation. It launched alongside the Qwen-Audio 3.1 speech and image model updates.
For buildersTest it against a live bilingual call or meeting feed and compare latency and fluency with a streaming translation API before committing.
Alibaba previewed the Qwen4 family at Apsara 2026, naming Qwen4-Max, Qwen4-Flash, Qwen4-Plus, and Qwen4-27B. The models are still in training with no release date, specs, or weights published yet, and the Qwen 4.5 and Qwen 5 series are planned at 5 to 10 trillion parameters.
For buildersKeep building on Qwen3.8-Max for now and watch the 27B tier if you want a locally deployable multimodal option when Qwen4 ships.
xAI released Grok 4.7, a coding and knowledge-work model listed at the same API price as Grok 4.6: $2 per million input tokens and $6 per million output tokens. It is available in Cursor, Grok Build, and the xAI API.
For buildersRun a long coding or document task on Grok 4.7 in Cursor or the API at Grok 4.6 pricing and compare quality with the model you currently use.
Qwen released Qwen-Image-2.1, an open-source image model that combines text-to-image generation and image editing in a single pipeline with native transparent image support and multi-image editing.
For buildersTry generating a product image with an alpha channel in a no-code design tool, then edit it locally using reference images in one model workflow.
Qwen released Qwen3.8-Omni-Flash, an omnimodal AI model with 1M-token context that processes text, image, audio, and video inputs with agentic task planning and tool use capabilities.
For buildersTest the model on a multi-step video editing or meeting summarization workflow to evaluate its omnimodal planning and tool use reach.
StepFun released Step 5 Preview, its flagship agentic model with 600B total parameters, 27B active parameters, and a 1M-token context window. It is available through the StepFun API, and open weights are now hosted in StepFun’s official Hugging Face repository.
For buildersEvaluate it for long-horizon coding and agentic workflows through the StepFun API. The 1M context window makes it suited for tasks that require sustained context across many turns.
xAI released Grok Voice Transcribe 2.0, a speech-to-text model available through its Speech-to-Text API for batch and streaming transcription. Company-listed pricing is unchanged from version 1.0: $0.10 per hour of audio for batch and $0.20 per hour for streaming.
For buildersSend a noisy support call or meeting recording through the Speech-to-Text API and compare speaker labels, timestamps, and formatted emails or phone numbers against your current transcriber.
Google introduced two live dialogue models, with a separate Extended Thinking option for more complex voice tasks.
For buildersPrototype a voice onboarding assistant that answers questions and walks new users through a multi-step setup.
TypeSafe AI released Jev, its first public System One model, in early access. It evaluates supplied state and returns typed decisions with probabilities instead of generating free-form text.
For buildersTry classifying support tickets or routing requests using predefined choices, with a review path for uncertain decisions.
DeepSeek released a new Flash model with native visual understanding and improvements to agent workflows.
For buildersTest screenshots and document images alongside the text-only tasks in your evaluation set.
OpenAI introduced an image model with sharper details, more precise editing, and faster generation.
For buildersUse repeated edits to evaluate whether a product image keeps its key details intact.
OpenAI released GPT-6 Astra, a new generation of model for demanding reasoning and computer-based work.
For buildersEvaluate a difficult multi-step task against explicit acceptance criteria before changing your default model.
Google announced 3.8 Flash with improvements to coding, reasoning, and multi-step agent workflows, alongside a separate Cyber variant.
For buildersEvaluate it on your app's real tasks: extracting data, routing support requests, or carrying out a sequence of tool calls.
Anthropic introduced Fable 5.1 for coding and longer knowledge-work tasks, with availability through paid Claude plans and its developer platform.
For buildersTry an end-to-end feature brief: ask it to inspect your app, implement the change, write checks, and explain the result.
Anthropic announced Mythos 5.1 with gains in cybersecurity and biology, retaining restricted access for vetted organizations.
For buildersCheck eligibility and deployment conditions before planning an integration.
Google launched a speech-to-text model designed to turn raw audio into polished, formatted text through the Gemini API.
For buildersAdd voice notes to a CRM or turn recorded customer interviews into searchable transcripts for your team.
A new hybrid architecture added native visual capabilities.
For buildersEvaluate coding with screenshot feedback.
Z.ai documented its updated coding model on the platform.
For buildersCompare difficult debugging tasks.
Grok 4.6 introduced improvements for longer-running agents and interactive visual work.
For buildersTest an agent on both the implementation and the rendered result of an interface change.
MiniMax launched a multimodal generation model producing video with native stereo audio.
For buildersPrototype a scene with coordinated visual action and environmental sound.
ByteDance introduced longer video generation with expanded reference and editing capabilities.
For buildersTry a continuous sequence with a clear beginning, middle, and end.
Opus 5 launched with improvements in coding and knowledge work, positioned below Fable in price.
For buildersBenchmark the full workflow, including tool calls and retries, when comparing model costs.
V8.2 updated aesthetics, personalization, and image editing.
For buildersTest a personalized visual style across a campaign.
Qwen introduced the third generation of its foundational image-generation model family.
For buildersCompare typography, instruction following, and edit consistency on the same creative brief.
Grok 4.5 launched for coding, agentic tasks, and knowledge work, including API access.
For buildersUse realistic engineering tasks to compare completion quality and total token use.
Kimi K3 launched with native vision and a one-million-token context window. Hosted access came first, with model weights scheduled afterward.
For buildersTest a large-repository coding task with screenshots and written requirements together.
OpenAI introduced a new generation of voice models for more natural spoken interaction.
For buildersTest interruptions and turn-taking as well as transcription accuracy in a voice assistant.
ByteDance introduced an image generation model emphasizing reasoning and design-oriented creation.
For buildersEvaluate a complete visual layout against a written design brief.
Sonnet 5 launched with improved planning, tool use, coding, and professional work at the Sonnet price tier.
For buildersMeasure whether a smaller model can complete the same workflow reliably before moving to a higher tier.
Google introduced a faster, lower-cost Gemini image model for generation and editing.
For buildersUse a lower-cost image model to explore many directions before refining a final asset.
OpenAI introduced the GPT-5.6 family through a limited preview: Sol, balanced Terra, and lower-cost Luna.
For buildersThe family offered separate capability and cost tiers for different workflow steps.
GLM-5.2 added a one-million-token context window.
For buildersTest document-heavy engineering tasks.
Google introduced a Gemma-based model using diffusion techniques to accelerate text generation.
For buildersMeasure both output quality and interactive latency when comparing generation architectures.
Fable 5 introduced Anthropic’s fifth generation for coding and knowledge work. Access was subsequently suspended and restored.
For buildersFor extended agent tasks, review intermediate progress and verify claimed completion.
Mythos 5 launched for a limited set of vetted partners, with a focus on cybersecurity and scientific research.
For buildersThis is a restricted-access research model, not a general-purpose public API choice.
NVIDIA released the largest model in its Nemotron 3 family.
For buildersCompare demanding reasoning workloads with the smaller Nemotron models.
A 12B multimodal model joined Gemma 4, positioned for laptop deployment.
For buildersA laptop-based assistant can process local documents without sending every input to a hosted model.
M3 introduced native multimodality and support for a one-million-token context window alongside coding improvements.
For buildersEvaluate visual inputs and a large codebase in the same workflow.
Anthropic released Claude Opus 4.8, an update to its Opus model family with a focus on more capable coding work.
For buildersTry it on an existing app's stubborn bug with a clear reproduction case, then review and test the proposed fix.
Mistral introduced a 128B open-weight model for longer coding and productivity workflows.
For buildersTest whether an agent maintains context and follows the brief across a full implementation task.
Google introduced Gemini 3.5 Flash for agentic and coding tasks, with access through the Gemini app, AI Studio, and its API.
For buildersBuild a first AI-powered utility in AI Studio, then test whether the model handles your users' inputs reliably.
DeepSeek released open-weight V4 preview models with one-million-token context and thinking/non-thinking modes.
For buildersCompare the Pro and Flash variants on your own long-input tasks.
OpenAI released GPT-5.5 for coding, research, data analysis, and work across computer tools. API availability followed on April 24.
For buildersTurn a messy brief and spreadsheet into a small internal tool, then check the calculations and behavior against the original data.
A new image model introduced stronger instruction following and generation of complex visual layouts.
For buildersTest diagrams and designs that require accurate text and consistent spatial relationships.
Moonshot released an updated open model focused on coding and sustained agent workflows.
For buildersEvaluate changes spanning multiple files rather than judging only single code snippets.
Anthropic introduced an Opus update with stronger coding, vision, and complex multi-step task performance.
For buildersFor visual app work, test whether the model follows both functional requirements and design references.
A specialized reasoning model for biology and drug-discovery research launched through a restricted research preview.
For buildersSpecialized models require domain-specific evaluation and an appropriate access program.
A new text-to-speech model expanded control over expressive generated speech.
For buildersPrototype spoken instructions and compare clarity across different speaking styles.
V8.1 improved speed and introduced direct 2K image generation.
For buildersCompare native high-resolution outputs with upscaled images.
GLM-5.1 targeted sustained agentic engineering.
For buildersMeasure progress across long coding sessions.
Gemma 4 introduced open models focused on reasoning and agent workflows across different hardware sizes.
For buildersTest an offline assistant on the device class your users actually own.
Google introduced an audio model for faster live conversations with longer conversational context.
For buildersTest a spoken support flow across interruptions and follow-up questions.
Adobe announced general availability of Firefly Image Model 5 after its preview phase.
For buildersEvaluate image generation and editing together within a design workflow.
MiniMax introduced a model focused on software engineering, professional tasks, and coordinated agent work.
For buildersTest an engineering task that requires planning, implementation, and review.
OpenAI added smaller GPT-5.4 models aimed at fast, cost-sensitive workloads.
For buildersEvaluate separate models for routine classification and more demanding coding requests.
An early-access V8 alpha launched on the Midjourney website.
For buildersTreat alpha results as an evolving model baseline.
Mistral Small 4 combined reasoning, multimodal understanding, and agentic coding in one model.
For buildersA unified model can simplify an app that mixes screenshots, text, and code.
GPT-5.4 brought reasoning, coding, and professional tool workflows together, with a Pro variant for harder tasks.
For buildersCompare the model on an end-to-end workflow spanning code, documents, and data.
Flash-Lite introduced a lower-cost option in the Gemini 3.1 family.
For buildersConsider lightweight models for high-volume extraction and classification.
Google introduced an image model combining capabilities from Nano Banana Pro with Flash-class speed.
For buildersCompare image quality and turnaround time for batches of marketing variations.
Google introduced Gemini 3.1 Pro for more complex reasoning and multimodal tasks.
For buildersEvaluate the model on realistic combinations of documents, images, and instructions.
Sonnet 4.6 improved coding, computer use, and planning, with a one-million-token context window in beta.
For buildersCompare a Sonnet-tier model with a larger model on your actual automation workload.
Qwen3.5 introduced an open-weight native vision-language model with reasoning and agent capabilities.
For buildersEvaluate multilingual and visual inputs together when your users work across languages.
Z.ai documented GLM-5 availability for complex engineering tasks.
For buildersEvaluate reasoning across multiple services.
A smaller coding model debuted in research preview, optimized for very low-latency inference on Cerebras hardware.
For buildersFast inference matters for interactive edits where a developer reviews each short change.
MiniMax released a model focused on coding, tool use, and practical productivity tasks.
For buildersMeasure success on complete workflows with a clear finished output.
ByteDance released a multimodal video model accepting text, image, audio, and video references.
For buildersTry guiding a video with a reference clip and a separate audio track.
Opus 4.6 improved coding and longer agent tasks, with a one-million-token context window in beta.
For buildersLarge-codebase work needs checks for both retrieval accuracy and correctness of the final change.
OpenAI introduced a coding model combining software-engineering ability with broader reasoning and computer-based work.
For buildersGive a coding agent a reproducible issue and verify both its fix and its tests.
Kuaishou launched Video 3.0, Video 3.0 Omni, Image 3.0, and Image 3.0 Omni with stronger consistency and narrative control.
For buildersTest whether references preserve characters across a multi-shot sequence.
Mistral introduced new batch and streaming transcription models.
For buildersChoose batch processing for recordings and streaming transcription for live captions.
Kimi K2.5 combined visual understanding, coding, and agent capabilities.
For buildersTry translating a screenshot into a working interface and reviewing the result visually.
M2.1 expanded multilingual programming and development capabilities.
For buildersEvaluate code quality across the languages used in your application.
GLM-4.7 improved coding and agent task execution.
For buildersCompare complete development tasks.
GPT-5.2-Codex launched in Codex for paid ChatGPT users, with API access planned afterward.
For buildersTest a migration or refactor that requires maintaining context across a long session.
Google expanded Gemini 3 with a faster, lower-cost Flash model for reasoning and multimodal tasks.
For buildersUse a fast model for interactive tasks where waiting time affects the user experience.
OpenAI released an image model with faster generation and more precise edits that preserve existing details.
For buildersEvaluate a series of edits for consistency, rather than judging only the first image.
NVIDIA released Nano as the first available model in its Nemotron 3 family, with larger models announced for later release.
For buildersEvaluate an open model for a focused agent workload on your own infrastructure.
The GPT-5.2 series launched for professional work, including documents, spreadsheets, coding, and longer tool workflows.
For buildersTest complete deliverables, such as an editable report and its calculations, rather than isolated answers.
Mistral released the second Devstral coding generation in 123B and 24B sizes.
For buildersCheck each variant’s license and hardware requirements before choosing a deployment.
Amazon introduced Nova 2 Lite and a preview of Nova 2 Pro with expanded reasoning capabilities.
For buildersEvaluate everyday automation separately from complex reasoning workloads.
Amazon released its second-generation real-time conversational speech model.
For buildersPrototype a spoken assistant and measure its handling of interruptions and follow-ups.
Mistral released Large 3 alongside smaller 3B, 8B, and 14B models under Apache 2.0.
For buildersA family of model sizes supports deployment from smaller devices to larger server workloads.
DeepSeek released reasoning-first models for agent workflows, with a higher-reasoning Speciale variant on a temporary endpoint.
For buildersDifferent reasoning tiers should be compared on successful task completion and cost.
Runway announced a video model with improved motion quality, prompt adherence, and visual fidelity.
For buildersCompare complex motion prompts and multi-part scene descriptions.
Black Forest Labs introduced an image model for generation and editing with multiple visual references.
For buildersUse reference images to test product, character, and style consistency across a set of assets.
Opus 4.5 launched with stronger coding, agent, and computer-use capabilities at a lower Opus price.
For buildersCompare completion quality and the number of retries, not just a model’s per-token price.
Nano Banana Pro brought Gemini 3 reasoning to image generation, including better text rendering and grounded visual content.
For buildersTest diagrams, campaign assets, and localized layouts with readable text.
OpenAI introduced a coding model designed for longer engineering tasks and more efficient reasoning.
For buildersCompare sustained debugging across several iterations.
xAI released a fast model for tool-using agents, alongside its Agent Tools API.
For buildersMeasure the latency of the whole tool workflow, not just the first generated token.
Gemini 3 Pro launched with improvements in reasoning, multimodal understanding, and agentic coding.
For buildersTest visual reasoning and code generation together when building interfaces from references.
An updated Grok model focused on conversational quality, nuanced instructions, and collaborative interactions.
For buildersEvaluate tone and instruction following using the conversations your users actually have.
The GPT-5.1 update introduced more conversational responses and more adaptive use of reasoning time.
For buildersA task router can reserve longer thinking for the inputs that benefit from it.
ElevenLabs introduced a streaming speech-to-text model for low-latency transcription.
For buildersEvaluate live captions with background noise, interruptions, and specialized vocabulary.
Moonshot introduced a thinking model designed for reasoning with repeated tool use.
For buildersTest multi-step research tasks that require gathering and reconciling information.
MiniMax released an open model focused on coding and agent tool use.
For buildersCompare repository tasks that require both code changes and command execution.
Anthropic released a new small model for coding, computer use, and responsive assistants.
For buildersCompare task completion and response time against larger models on short coding jobs.
Veo 3.1 improved realism, audio, and creative control in video generation.
For buildersEvaluate whether a generated sequence follows the intended story and preserves its subjects.
GLM-4.6 expanded coding capabilities and context length.
For buildersTest changes across a larger repository.
Sora 2 introduced video with synchronized dialogue and sound effects. The Sora product was later discontinued in April 2026.
For buildersA historical example of integrating video and audio generation into one model.
Sonnet 4.5 launched with advances in coding, computer use, and complex agent workflows.
For buildersEvaluate long-running agents on completion quality, recovery from errors, and total cost.
An experimental model introduced DeepSeek Sparse Attention for more efficient long-context processing.
For buildersLong-context efficiency matters for repeated analysis of large documents or codebases.
Luma unveiled a reasoning-driven video model with creative controls and an HDR production pipeline.
For buildersExplore generated shots that need controlled transitions and consistent visual direction.
OpenAI released a GPT-5 model specialized for agentic software development.
For buildersEvaluate a complete repository change with implementation and test execution.
An updated Kimi K2 improved agentic coding and expanded its context window to 256K.
For buildersUse a multi-file app change to assess whether longer context translates into better implementation.
Google released an image generation and editing model with stronger character consistency across edits.
For buildersTry changing backgrounds while preserving a product or character across several images.
DeepSeek introduced a hybrid model combining thinking and non-thinking modes, with stronger agent capabilities.
For buildersUse reasoning selectively for the steps that require planning or difficult decisions.
GPT-5 introduced a system combining fast responses and deeper reasoning, with improvements across coding and multimodal tasks.
For buildersEvaluate both quick responses and longer reasoning on the same realistic set of app tasks.
An Opus update improved agentic coding, reasoning, and detail tracking during research and analysis.
For buildersMulti-file refactoring is a useful test of whether a model makes precise changes.
OpenAI released two open-weight reasoning models under the Apache 2.0 license.
For buildersOpen weights enable local deployment and customization when the hardware and license fit your project.
Z.ai launched language models for reasoning, coding, and agents.
For buildersEvaluate multi-step tasks using tools.
Moonshot released open-weight mixture-of-experts models focused on knowledge, coding, and agentic tool use.
For buildersEvaluate how a model chooses and sequences tools, as well as the final answer.
Grok 4 launched with reasoning, native tool use, and real-time search integration; Heavy provided a higher-compute option.
For buildersGround current-information tasks in retrieved sources and verify the resulting citations.
Alibaba previewed a model combining visual understanding with image generation and editing.
For buildersExplore edits described through both a reference image and written instructions.
Google added a lower-cost, faster model to the Gemini 2.5 family in preview.
For buildersTest high-volume document classification and extraction with representative inputs.
Mistral introduced its first reasoning-model family, emphasizing multilingual reasoning.
For buildersEvaluate complex reasoning in the languages your app supports.
An expressive text-to-speech model debuted in alpha with audio tags and multi-speaker dialogue.
For buildersDirect delivery and emotion for a narrated lesson, then listen for clarity and consistency.
Kontext introduced image generation and editing conditioned on text and existing images.
For buildersUse a reference image to preserve identity while changing a scene.
An updated R1 checkpoint improved reasoning and added support for structured outputs and function calling.
For buildersEvaluate a reasoning workflow that must return valid structured data.
Claude 4 launched with stronger coding, reasoning, and longer-running agent capabilities.
For buildersUse repository-level tasks to evaluate whether a coding agent preserves existing behavior.
Mistral and All Hands AI released an open model for agentic software-engineering tasks.
For buildersRepository navigation and tool use matter when a model must fix a real software issue.
Google introduced its next image-generation model with improvements in visual detail and text rendering.
For buildersCompare image models on the text, layout, and brand constraints in your actual creative brief.
Google introduced Veo 3 for video generation with native audio.
For buildersPrototype short scenes with speech and environmental sound before committing to a full production.
Mistral introduced a model positioned to balance capability, efficiency, and enterprise deployment needs.
For buildersInclude latency and hosting requirements in your model comparison.
Qwen3 introduced open-weight dense and mixture-of-experts models with thinking and non-thinking modes.
For buildersTest whether extra reasoning improves accuracy enough to justify added latency.
Adobe released Image 4 and Image 4 Ultra for image generation, with Ultra focused on complex scenes and detail.
For buildersCompare simpler asset production with compositions involving several subjects.
A preview of Gemini 2.5 Flash introduced controllable thinking budgets while prioritizing speed and cost.
For buildersThinking budgets let you balance latency and reasoning effort for each type of request.
The new reasoning models combined extended thinking with tool use and visual analysis.
For buildersA reasoning workflow can inspect an image, run calculations, and gather evidence before returning an answer.
Three API models launched with improved coding and instruction following, and context windows up to one million tokens.
For buildersDifferent model sizes let an app route simple extraction and complex code work to different cost tiers.
Meta released two open-weight, natively multimodal mixture-of-experts models. Behemoth was described but not released.
For buildersMultimodal open weights offer another route to custom document and image-understanding systems.
V7 improved image quality and prompt interpretation.
For buildersExplore a consistent set of visual concepts from reference images.
Runway introduced a video model focused on consistent characters, objects, and environments across shots.
For buildersBuild a short storyboard using the same subject across several scenes.
Google introduced the first Gemini 2.5 model as an experimental reasoning model with stronger coding capabilities.
For buildersA useful milestone for comparing multimodal reasoning with earlier Gemini generations.
OpenAI introduced native image generation with improved prompt following, text rendering, and image-based editing.
For buildersCreate a visual from a brief, then refine it using the previous output as context.
OpenAI released GPT-4o Transcribe, GPT-4o mini Transcribe, and GPT-4o mini TTS for speech recognition and generation.
For buildersCombine transcription and speech generation to prototype a voice-based support assistant.
Cohere released Command A for enterprise language and agent workflows.
For buildersEvaluate grounded answers and tool use against internal business documents.
Google released a new open-model family with multilingual and multimodal capabilities.
For buildersOn-device document and image understanding can reduce reliance on hosted APIs.
OpenAI released a research preview emphasizing conversational quality, broad knowledge, and instruction following.
For buildersThis release illustrates that conversational quality and explicit reasoning are different model strengths.
ElevenLabs released its first speech-to-text model, with multilingual transcription, timestamps, and speaker labels.
For buildersCreate searchable meeting or interview transcripts with links back to the original recording.
Anthropic introduced a hybrid reasoning model that could respond immediately or use extended thinking.
For buildersA single model can support both quick assistance and more deliberate technical work.
xAI detailed its Grok 3 beta models and their reasoning capabilities, including a smaller reasoning model.
For buildersCompare general knowledge tasks with focused mathematical reasoning.
A small reasoning model arrived with a focus on science, mathematics, and coding.
For buildersReasoning effort can be adjusted to match the difficulty of a technical task.
DeepSeek released its reasoning model and six smaller distilled models with open weights.
For buildersCompare reasoning quality across model sizes before choosing a deployment approach.
Luma launched Ray2 on a new multimodal architecture for video generation.
For buildersPrototype short scenes and check whether movement remains plausible throughout the clip.
DeepSeek released a large mixture-of-experts language model with open weights.
For buildersA milestone for comparing hosted APIs with independently deployable language models.
Google introduced Veo 2 with improved realism and cinematic control through a limited VideoFX rollout.
For buildersCompare motion and camera behavior over a series of generated clips.
Microsoft introduced a 14-billion-parameter language model emphasizing reasoning, initially through Azure AI Foundry.
For buildersCompare a compact model on focused mathematical and structured reasoning tasks.
Gemini 2.0 began with an experimental Flash release and expanded native multimodal and tool-use capabilities.
For buildersA reference point for moving from chat responses to models that participate in tool workflows.
A faster version of Sora became available for video generation after the earlier research preview.
For buildersThis historical release marks the move from a research demonstration to a usable video model.
The full o1 model arrived in ChatGPT alongside a higher-compute o1 pro mode.
For buildersCompare reasoning depth against latency when designing a multi-step assistant.
Amazon launched Nova understanding models alongside Canvas for images and Reel for video in Bedrock.
For buildersCompare text, multimodal, and media tasks within an existing AWS application.
An updated Sonnet model improved coding and introduced experimental computer use through the API.
For buildersComputer-use agents need evaluation on real screen flows, including unexpected states.
Stability AI released customizable image models, with the Medium variant following on October 29.
For buildersChoose a model variant against your available hardware and required image quality.
Meta released small text models for edge devices and larger models with vision capabilities.
For buildersMatch model size and modality to the device and task instead of defaulting to the largest model.
Alibaba’s Qwen team released a broad family of open-weight language models, with coding and mathematics variants.
For buildersChoose the smallest model that meets the requirements of your language or code task.
OpenAI released a smaller reasoning model focused on tasks such as coding and mathematics.
For buildersA lower-cost reasoning tier offered another option for technical tasks.
An early reasoning model became available in ChatGPT and to eligible API users, using more computation before answering.
For buildersReasoning models introduced a new tradeoff between response speed and work on difficult problems.
xAI introduced a new model generation with improved text and visual understanding.
For buildersCompare general conversation and image understanding as separate capabilities.
Black Forest Labs introduced its first FLUX image generation models.
For buildersCompare prompt adherence and visual consistency across a small set of brand assets.
V6.1 refined image details, textures, and generation speed.
For buildersCheck fine details when comparing visual directions.
The second Mistral Large generation introduced 128K context and stronger multilingual and coding capabilities.
For buildersEvaluate retrieval across long inputs before relying on a large advertised context window.
Llama 3.1 expanded context to 128K, added language support, and introduced the 405B model.
For buildersCompare the infrastructure costs of larger open-weight models with a managed API.
A smaller, lower-cost model launched with text and vision support.
For buildersSmall models can make classification, extraction, and repeated background tasks more economical.
Gemma 2 launched in 9B and 27B sizes with improvements to model quality and inference efficiency.
For buildersSelect a model size against your available memory and expected request volume.
The first Claude 3.5 model introduced improvements in coding, visual reasoning, and speed.
For buildersA milestone for generating interfaces and reviewing screenshots alongside code.
Runway announced its next-generation video model with improved motion, consistency, and temporal control ahead of rollout.
For buildersUse the examples to plan short generated scenes with deliberate camera movement.
Alibaba released five Qwen2 model sizes, including dense and mixture-of-experts models.
For buildersUse the size range to compare hosting requirements for a multilingual assistant.
Mistral introduced its first dedicated code model, trained across more than 80 programming languages.
For buildersCode completion and repository-level agents are different tasks and should be evaluated separately.
Google introduced a lighter Gemini model designed for speed and efficiency at scale.
For buildersLatency-sensitive summarization and extraction often benefit from a smaller model tier.
Google introduced its third-generation image model, initially available to selected creators.
For buildersCompare fine details and prompt interpretation in generated visual concepts.
Google introduced Veo for high-definition video generation with limited creator access and a waitlist.
For buildersExplore how camera and cinematic language translate into generated video.
OpenAI introduced its omni model for text, vision, and audio. Text and image inputs launched first; other modalities followed.
For buildersA useful reference point for the shift toward assistants that understand screenshots and spoken input.
Adobe introduced its third-generation image model in beta with improved quality and creative control.
For buildersCompare product compositions and lighting against earlier image model generations.
Microsoft released Phi-3 mini, a compact language model designed for constrained compute environments.
For buildersTest whether a small model can handle a focused task close to the user device.
Meta introduced the next generation of Llama with improved language and instruction-following capabilities.
For buildersUse a consistent evaluation set when comparing hosted and self-hosted models.
Cohere introduced a larger RAG-oriented model for enterprise workloads.
For buildersCompare answer grounding and tool-use reliability on representative business documents.
xAI announced Grok-1.5 with improved reasoning and a 128K context window ahead of early-user rollout.
For buildersCompare long-document tasks with the preceding Grok generation.
Haiku joined the Claude 3 family as its fast, lower-cost model.
For buildersA useful historical example of routing straightforward requests to a smaller model.
Cohere introduced Command R for retrieval-augmented generation in production applications.
For buildersBuild answers around retrieved documents and verify that citations support the claims.
Opus and Sonnet launched as the first available Claude 3 models, with vision support and different capability tiers.
For buildersCompare a larger and smaller model on the same document-understanding tasks.
Mistral introduced its flagship model for multilingual reasoning, text transformation, and code generation.
For buildersTest multilingual prompts using native-language examples from your actual users.
Google introduced Gemma, a family of lightweight open models related to its Gemini research.
For buildersSmaller downloadable models can support experimentation on hardware you control.
Google introduced a Gemini 1.5 Pro preview with a breakthrough in long-context multimodal understanding.
For buildersLarge document and video collections became practical inputs for a single model request.
V6 improved long-prompt interpretation and image coherence.
For buildersCompare detailed composition prompts against earlier generations.
Mistral announced an open-weight sparse mixture-of-experts model under Apache 2.0.
For buildersMixture-of-experts models balance total model capacity against the computation used per token.
Google introduced the Gemini family. Pro and Nano began rolling out first, while Ultra was announced for a later release.
For buildersThe family established separate model sizes for cloud applications and on-device tasks.
Claude 2.1 expanded context to 200K tokens and added a tool-use beta.
For buildersLong context and external tools enable assistants grounded in a company’s documents.
GPT-4 Turbo launched in preview with a 128K context window and lower API prices.
For buildersLonger context made it possible to supply much larger documents in a single request.
xAI announced the Grok-1 language model powering the early Grok beta.
For buildersA historical starting point for the Grok family of conversational models.
Following its research preview, DALL-E 3 became available to ChatGPT Plus and Enterprise users, with better handling of detailed prompts and text.
For buildersUse conversational revisions to refine an image concept over several steps.
Mistral released a compact language model under the Apache 2.0 license.
For buildersA small self-hosted model can be a useful baseline for a narrowly defined text task.
Stability AI released SDXL 1.0, a higher-resolution text-to-image model with downloadable weights.
For buildersA key release for custom image workflows built around local generation and fine-tuning.
Meta released Llama 2 models for research and commercial use under its community license.
For buildersDownloadable weights opened more choices for hosting and customizing a conversational model.
Claude 2 launched with improved coding, mathematics, reasoning, and longer responses, alongside a public web beta.
For buildersLong-document assistants became a practical use case for the Claude family.
Google introduced a language model family with improved multilingual, reasoning, and coding capabilities.
For buildersA useful milestone in the evolution of multilingual assistants before Gemini.
Adobe introduced the first Firefly model in beta for image generation and text effects.
For buildersA milestone in bringing generative image models into established design workflows.
Anthropic introduced its first broadly offered Claude models, including a faster, lighter Instant version.
For buildersAn early choice of model tiers for summarization, writing, and conversational applications.
OpenAI introduced GPT-4, with improved reasoning and a multimodal architecture accepting text and images.
For buildersUse this release to trace the move from text assistants toward visual reasoning.
Meta introduced a family of foundation language models for research, ranging from 7B to 65B parameters.
For buildersA foundational milestone in the development of downloadable language models.
The GPT-3.5-based conversational model debuted as a free research preview in ChatGPT.
For buildersA milestone in using natural-language instructions to draft, explain, and debug software.
Stable Diffusion 2.0 introduced a new text encoder, higher-resolution generation, and depth-guided image transformations.
For buildersExplore how preserving depth can help an image edit retain its original composition.
OpenAI released open-source speech-recognition models for multilingual transcription and translation into English.
For buildersTurn recorded interviews or lessons into searchable transcripts.
Stability AI, LMU Munich researchers, and Runway co-released Stable Diffusion for text-to-image generation.
For buildersOpen image-model weights enabled local generation and a large ecosystem of custom workflows.
Meta shared a large pretrained language model for research, alongside smaller models and training documentation.
For buildersA milestone in making large language model research more reproducible.
OpenAI began a limited research preview of its second-generation text-to-image model.
For buildersA historical milestone for generating and editing images with natural-language prompts.
A history of major AI model launches across language, coding, image, video, and speech, with official sources and practical notes for builders.
Major models from 2022 onward, including OpenAI, Anthropic, Google, Meta, DeepSeek, Alibaba, Mistral, and image, video, and speech providers. Same-day model variants may share an entry. Product features, integrations, minor checkpoints, and community fine-tunes are excluded.
Dates follow the linked official announcement or documented release milestone. Restricted previews and announcements ahead of rollout are labeled. Access can vary by account, region, or plan, and historical models may no longer be available. Summaries describe the announcement; project ideas are our editorial suggestions, not hands-on benchmark results.
Start with your task, budget, and the tools your app needs. Test a candidate model on realistic inputs and compare its accuracy, latency, and cost. A newer release is a reason to evaluate, not an automatic reason to switch.
Compare AI and no-code tools →Follow the related guides in the timeline, browse our tool directory, or choose a practical AI and no-code course to build a working project step by step.