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.
MiniMax lists community quantizations for running M3.1 Flash locally
MiniMax posted a roundup of community-made quantized versions of its 428B-parameter open-weight M3.1 Flash, aimed at people running it on their own hardware. MiniMax says lower-bit weights cut the memory needed to load the model and the weight data moved during decoding, and that making this useful takes careful calibration, efficient low-bit kernels and placing experts across GPU and system RAM.
Claude Mythos 5.5 is not announced, but Anthropic's website source mentions it
Anthropic has not announced a Claude Mythos 5.5. Several X accounts report that the HTML source of Anthropic's Claude Mythos page (anthropic.com/claude/mythos) names it.
Google announces Gemini 4 Carbon, its next frontier model, to cyber defenders first
Google announced Gemini 4 Carbon on October 28, 2026, calling it its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. It follows Gemini 4 Argon by two weeks. Sundar Pichai's post and Google's blog say it is not yet generally available: starting this week a subset of cyber defenders get access, with more trusted partners added through the Fairwind Program. Google says it will bring Carbon to developers, businesses and consumers "as soon as possible", and that it is taking part in the U.S. government's voluntary process for pre-release model testing.
OpenRouter's catalog gives Z.ai: GLM 4.7 an expiration date of December 31
OpenRouter's model catalog now shows an expiration date of December 31, 2026 for "Z.ai: GLM 4.7" (z-ai/glm-4.7), a model listed since December 22, 2025 with a 204,800-token context. The entry had no expiration date before. This is a catalog field only; the entry does not say why or what happens to the model after that date.OpenCode says its DeepSeek V4.1 Flash configs match DeepSeek's API, after reports the model feels dumber
OpenCode's founder (thdxr) wrote that OpenCode Go gets occasional reports that DeepSeek V4.1 Flash "feels dumber" for some people. He said he ran every benchmark he could find against the different configurations OpenCode Go serves, and that the results are in the same range.Rock Matrix open-sources EdgeRock-4B, a small reasoning model with a fixed-size KV cache
Rock Matrix released the weights and code of EdgeRock-4B, a 4B text-and-image reasoning model, on GitHub, Hugging Face and ModelScope, with NF4, INT8 and FP8 versions. A paper is on arXiv (2610.10118). The announcement was relayed on X by TestingCatalog on October 31.GPT-6 Astra for Law is not announced, but appears on OpenAI's pricing page and model list
OpenAI has not announced a legal version of GPT-6 Astra. Two X accounts report that it has shown up on OpenAI's own pages.Anthropic sets thousands of Claude agents to red-team training-data monitors for emergent misalignment
Anthropic published research on automating red-teaming for subtle emergent misalignment, saying red-teaming is vital to model safety testing but limited by researchers' time. Everything below is Anthropic's own account, from its post and its Alignment Science Blog write-up.
Claude Code 2.1.316 deprecates fast mode for Opus 4.8, with removal on November 30
Claude Code v2.1.316, released October 30, 2026, per its release notes:Artificial Analysis details AA-Omniscience v1.1, with a smaller dataset and a new public set
Artificial Analysis published the details of AA-Omniscience v1.1, its knowledge and hallucination benchmark, which it updated on October 27 as part of v5 of its Intelligence Index. The benchmark's Net Score rewards correct answers, penalizes wrong ones and treats abstaining as neutral.OpenAI releases GPT-6 Rosalind, a biology model with scientific tool use, to approved organizations
OpenAI's API changelog for October 30, 2026 lists the release of GPT-6 Rosalind (gpt-6-rosalind), which it describes as bringing specialized biological reasoning with scientific tool use to complex life sciences research. Access is limited to organizations approved for Rosalind Research through a trusted-access program, for internal research by authorized users.Gemini API's Computer Use tool is generally available
Google's Gemini API changelog says the Computer Use tool has moved to general availability. It works across both the Interactions API and the GenerateContent API.Anthropic confirms a Claude Code prompt change to stop Opus 5.5 recreating specific games
Anthropic has confirmed that it added two sentences to Claude Code's prompt on Opus 5.5 to steer the model away from recreating a specific existing game. Game Developer reported it on October 30.
Mercor's APEX-Agents leaderboard puts Gemini 4 Carbon first, the only model with a mean above 90%
Mercor, which publishes the APEX-Agents benchmark, says Gemini 4 Carbon debuts at #1 on its leaderboard. These are Mercor's own measurements. APEX-Agents tests long-horizon, multi-step tasks drawn from investment banking, management consulting and corporate law.OpenAI tightens ChatGPT's violence safeguards and limits weapons information for teens
OpenAI announced stronger violence-prevention measures in ChatGPT and a commitment to a public Community Safety Report. Everything below is OpenAI's account; it says the safeguards begin rolling out "in the next few weeks".Artificial Analysis launches a leaderboard for HLE-Diamond, the refined Humanity's Last Exam
Artificial Analysis launched its HLE-Diamond leaderboard, its own evaluation of the refined 1,000-question version of Humanity's Last Exam, as part of v5 of its Intelligence Index. CAIS and Scale AI released HLE-Diamond after a year of cleaning and refining the questions.