AI TrackerModels news,
for builders.
The latest models news from every AI provider: launches, price changes and retirements, summarized with what they mean for you.
PokeBot launches RoboAPI, an API that gives robot arms natural-language instructions
PokeBot introduced RoboAPI on October 29, powered by Poke-Aurora, which it calls its world-action model. It says compatible robots can take natural-language instructions through a simple API, with no task-specific training.Pika launches H3 Flex, a fast video model it calls the top open-weight text-to-video model on DesignArena
Pika introduced H3 Flex, which it says debuts as the #1 open-weight model on DesignArena's text-to-video board (Pika's claim). It describes the model as built for professional iteration: exportable audio stems, resolutions up to 4K and faster-than-realtime generation, with a 15-second clip and soundtrack in as little as 7 seconds.OpenAI updates its Model Spec around five core values
OpenAI published an updated Model Spec, its public description of how it wants its models to behave. The revision reorganizes the guidance around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. Everything below is OpenAI's account.OpenAI releases GPT-6.1 Terra, a cheaper computer-use model at half the price of GPT-5.6 Terra
OpenAI released GPT-6.1 Terra on October 26, 2026, describing it as a fast, capable computer-use model: it understands what is on screen, takes action across apps and writes naturally. It is rolling out to all Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and developers can use it in the API as gpt-6.1-terra. OpenAI says usage allowances in ChatGPT Work and Codex go further with it than with GPT-5.6.Phonic 1 debuts at #1 on Artificial Analysis's Speech to Speech Index
Artificial Analysis (a benchmark site, its own measurements) says Phonic 1 (High), the speech-to-speech model from Phonic, takes #1 of 22 on its Speech to Speech Index at 87.1, narrowly ahead of Grok Voice 3.0 (86.9). The model is available through the Phonic API at $0.07 per minute.Fish Audio releases Drama 3, which Artificial Analysis ranks #1 among open-weight speech models
Fish Audio released Drama 3, with weights on Hugging Face. Artificial Analysis (a benchmark site, its own measurements) says it is the #1 open-weights model on both of its speech arenas and #1 of all models on its Pronunciation Robustness benchmark.Mistral releases the weights of Mistral Large 4, a 1-trillion-parameter model, at about half the preview price
Mistral published the weights of an improved Mistral Large 4 ("Le Chonk"), its flagship natively multimodal model of 1 trillion parameters, which entered public preview earlier this month. Weights are on Hugging Face as Mistral-Large-4-1T-A52B and an NVFP4 version. Mistral calls it European-trained and the best open-weight model for coding. A commenter says the license is a modified MIT, not an OSI-approved one.Qwen 4 is not announced, but Qwen Code already lists qwen4 model ids
An X account reports that Qwen Code v0.26.3, released on October 28, includes a change (pull request #15762 in the official QwenLM/qwen-code repository) that treats every model whose id starts with qwen4 as having a 1M-token context by default. According to that account, the tests name qwen4-max, qwen4-plus, qwen4-coder-plus, the dated qwen4-max-2026-10-01 and qwen4.5-max. The account's post links the pull request and shows a screenshot.Artificial Analysis shows AA-Robotics v0.1, a benchmark of models controlling a robot arm
Artificial Analysis published AA-Robotics v0.1, which has models control a robot arm and shows the task from the model's point of view. In an example where the same robot writes "hello world", Artificial Analysis says Claude Opus 5.5 dabbed it out dot by dot, with 3,270 arm moves over 34 minutes, while GPT-6 Astra wrote in continuous strokes, with 514 moves in 13 minutes.Particle-1, a model post-trained on DeepSeek V4.1 Flash, is now on Vercel AI Gateway
Vercel says Particle-1 (particle/particle-1) is now available on AI Gateway, its unified endpoint for model providers. Vercel says the model is post-trained on DeepSeek V4.1 Flash and completes engineering tasks with fewer tokens, and suggests it for debugging, planning code changes and breaking down multi-step problems.Artificial Analysis publishes AA-PersonalityBench: GPT-6 and Grok models are the most direct
Artificial Analysis published AA-PersonalityBench, a benchmark of model personality traits, with a results page, a methodology page and a tool for building your own personality ranking. The scores are the site's own.
ElevenLabs introduces Cue, a forced alignment model
ElevenLabs research introduced Cue, a forced alignment model: the step that turns a transcript into timestamps, used for subtitles, dubbing and clickable transcripts. It places every character on the audio timeline.Rime launches Prism, a speech-to-speech model for voice agents
Rime launched Prism, a speech-to-speech model for AI voice agents. The figures below are Rime's own claims, not independently measured.Artificial Analysis launches AA-Video-I2V v2.1, a benchmark for image-to-video models
Artificial Analysis launched AA-Video-I2V v2.1 and AA-Video-I2V-Silent v2.1, a benchmark for image-to-video models. Each prompt pairs a 1080p starting image with a prompt written for it and is tagged by use case, capability and visual style; models are judged at up to 1080p and ranked across 10 use cases and 10 capabilities. The overall ranking samples evenly across them. The results are Artificial Analysis's own.
GitHub Copilot Fusion, which routes and orchestrates models for each task, is generally available
GitHub says Copilot Fusion is generally available. Instead of choosing a model, a user picks Copilot Fusion in the model picker and an effort level (Standard, High or Max); Copilot decides which models to use and how they work together. It combines the multi-model orchestration of the earlier Project HydraFusion preview with Copilot's auto model selection. Workflows include a single model, a cascade (an efficient model drafts and a quality check decides whether to escalate) and critique (one model drafts, another reviews, the first revises).
OpenBMB releases MiniCPM-OCR, a 1B document parser, with a 2B model that scores OCR results
OpenBMB showed MiniCPM-OCR on X on October 28, running locally on an iPhone and an iPad. It is released with open weights and can run with llama.cpp, vLLM, SGLang or Ollama; fine-tuning code is open source, and OpenBMB also offers a demo and an API.AssemblyAI launches its first text-to-speech model, AssemblyAI-TTS-1
AssemblyAI launched AssemblyAI-TTS-1, its first text-to-speech model, built for voice agents. AssemblyAI says it streams in 90 ms at $15 per million characters and calls it the best TTS model you can run at scale on accuracy, naturalness and cost together. OpenRouter says the model is live there, with 4,000+ voices served from the US or EU.
NVIDIA releases open-weight Nemotron 3.5 Transcribe, which ties for #1 among open-weight speech-to-text models on Artificial Analysis
NVIDIA's Nemotron 3.5 Transcribe is a new open-weight, promptable speech-to-text model: a 30B mixture-of-experts with about 4B active parameters. NVIDIA says one checkpoint handles transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding, and that the weights, data and recipes are open. It is available on Hugging Face, with a live demo.Mistral opens a Responses API in public preview and serves DeepSeek V4.1 Flash and GLM 5.3 Flash
Mistral announced on October 28 that its Responses API, made for building agents on Mistral, is in public preview. According to Mistral, each response returns the model's answer, its reasoning and its tool calls as separate items, so an app can handle every step more easily than with Chat Completions. It works with all chat models hosted by Mistral, including Large 4. The documentation is on Mistral's Studio docs site.