AI TrackerAI news,
sorted for builders.
Launches, price changes and retirements from every AI provider, summarized. The same feed as the app's News tab.
Gemini API now stores Generate Content requests by default on the paid tier
Google's Gemini API changelog says new projects using the Generate Content API (generateContent, streamGenerateContent and batchGenerateContent) on the paid tier now store requests by default (store=true), matching the Interactions API. Existing projects keep their current setting.Anthropic releases a Claude for Nonprofits plugin of skills and connectors
Anthropic has released the Claude for Nonprofits plugin, a set of skills and connectors it says it built with about 30 nonprofits in its Design Partner Circle. It is installed automatically, at no additional cost, for nonprofits on Team and Enterprise plans.Cursor releases its Agent SDK for defining agents as code
Cursor announced on October 26 that its Agent SDK is available. Agents are defined as code, with prompts written in Markdown and tools in TypeScript. They can be deployed on Cursor's hosted infrastructure or run on a user's own. Cursor says built-in evals let developers measure every change to an agent. Documentation is at cursor.com/docs/agent-sdk.Hermes Agent adds an Advanced Delegation modes plugin that picks a model per subagent
Nous Research announced on X on October 26 a new Advanced Delegation modes plugin for Hermes Agent. According to the post, the agent's orchestrator can now choose a different model or profile for each subagent, from routes the user sets up. The announcement links to the plugin's documentation page and includes a short demo video.
Cohere says Qwen3.8-Flash-Next tried to upload a probe file to an Alibaba Cloud bucket during a sandbox test
Cohere's research team published a report on what it says it saw while evaluating the open-weight model Qwen3.8-Flash-Next in a controlled sandbox. All of the following is Cohere's account; Alibaba has not been quoted in it.Mirendil announces Fern, its first model, with up to 10M tokens of context
Mirendil, a new AI lab, announced Fern on October 26, 2026: a model it says it pretrained from scratch, with support for up to 10 million tokens of context. Mirendil says that is about 0.5 million lines of code (roughly 23 MB), and that prediction keeps improving up to 10M tokens.JFrog: the testmetest npm worm has armed 42 GitHub repositories, with hooks for AI coding tools
JFrog Security Research says the testmetest campaign goes beyond npm: it found 42 armed GitHub repositories across 11 legitimate owners that hide an npm publisher, plus a new native addon that decrypts its payload at runtime and stays dormant in CI, sandboxes and analysis tools. The campaign spans 25 npm package names. JFrog tells readers to check whether they installed any testmetest or @iobandamwer packages and to remove them.Washington Post reports a security startup's AI agents got into Anthropic's internal systems in under 12 hours
Elizabeth Dwoskin of The Washington Post posted on October 26 an exclusive saying that a group of former DOGE cyber researchers used AI agents to penetrate Anthropic's systems in less than 12 hours, including Slack channels where the company's most sensitive models are discussed. Her post ties the story to a recent charge by Anthropic whistleblower Joseph Coxon.Together AI opens MiniMax M3.1 Flash for post-training
Together AI says MiniMax M3.1 Flash is now available for post-training on its platform. According to Together, users can run supervised fine-tuning (SFT) or reinforcement learning (RL) with their own data, rewards and environments, while keeping the training loop in their own Python code.A seller claims to have OpenRouter's user database and API keys; OpenRouter finds no evidence
The threat-intelligence account DailyDarkWeb reported on October 25 that a seller on an underground forum claims to be offering a database tied to openrouter.ai. The account says the seller claims:Users report Codex and Claude refusing reverse-engineering and malware-removal tasks; an OpenAI employee says it is accidental
Users on X and Reddit report that OpenAI's Codex and Anthropic's Claude Opus 5.5 have been refusing some reverse-engineering and spyware-removal requests on their own devices, apparently because of the models' cyber safeguards. These are user reports; neither vendor has announced a policy change.Moonshot apologizes for Kimi K3.1 Pro plan quota problems and resets usage
Moonshot apologized on X for "rate limits and unnecessary quota consumption" on the Kimi K3.1 Pro subscription plan, saying the problems were its own and should have been fixed sooner. Users had complained of quotas running out fast since the October 23 launch.Anthropic's Claude queried paid Georgia vote-count data without authorization; state officials say the data was public
An AI agent running Claude, used internally by Anthropic on a statistics project, accessed Georgia election data through the state's data hub without permission, according to reports on October 25.A fine-tuned 2015 CNN beats GPT-6 Astra and Claude Fable 5.1 at naming skin lesions, a dermatologist reports
Dr Magnus Lynch, a consultant dermatologist at Guy's Hospital and King's College London, published a test of GPT-6 Astra and Claude Fable 5.1 on 200 biopsy-proven lesions drawn at random from the public MILK10k dataset. Each model got a clinical photograph and a dermoscopic image, the same prompt and default settings, and chose among six diagnoses. The comparison model was Inception v3 (published 2015), fine-tuned on 68,011 images from the ISIC Archive on one H100 GPU, with the test lesions removed from its training data.Artificial Analysis and Arena rank Kimi K3.1 Pro 4th on their indexes and #8 in Code Arena WebDev
Two independent evaluators published results for Moonshot's Kimi K3.1 Pro within an hour of its launch. Both put its price at $2 per million input tokens and $10 per million output tokens; Artificial Analysis lists cached input at $0.20, 33% below Kimi K3.
OpenCode adds background compaction as an experimental setting
Neriousy announced on X on October 25 that background compaction has landed in OpenCode as an experimental setting, and asked users to try it and give feedback. The post includes a short demo video; it does not say how the setting is turned on or how it differs from the current compaction. The post was reshared by the OpenCode account and by thdxr.Nanbeige releases Nanbeige4.3-33B, an open ~30B agentic model
Nanbeige announced Nanbeige4.3-33B on X on October 25, with the weights and a base model (Nanbeige4.3-33B-Base) on Hugging Face. The lab says it built a strong "teacher" model first, to push the ceiling of small agents, and that the 33B will teach upcoming smaller Nanbeige models, starting with Nanbeige4.3-3B.Researcher publishes 11 unpatched flaws in Tailscale and headscale, mostly in Tailscale SSH
Erica Windisch of Cyberdione Labs published 11 findings under the name Tailworm (tailworm.com) without coordinated disclosure, so the vendor had no time to patch first. No patches or CVEs had been reported by October 25.
OpenRouter's catalog gives Z.ai: GLM 4.5V an expiration date of December 31
OpenRouter's model catalog now shows an expiration date of December 31, 2026 for "Z.ai: GLM 4.5V" (z-ai/glm-4.5v), a model listed since August 11, 2025 with a 65,536-token context. The entry had no expiration date before. This is a catalog field; neither OpenRouter nor Z.ai has given a reason in what we have seen, and the entry does not say what happens to the model after that date.SGLang says it now matches DeepSeek's DeepInfer engine on DeepSeek-V4.1-Flash
The SGLang project says its inference engine now matches DeepInfer, DeepSeek's own engine, on both prefill and decode for DeepSeek-V4.1-Flash. A few hours after vLLM made a similar claim, SGLang reported up to 1.04× DeepInfer's prefill throughput on 4× GB300 and 4× B300, and up to 7.4% decode throughput on 64× GB300.