EverMind AI launches Raven, a pre-alpha self-evolving multi-agent harness that builds DAG workflows for complex tasks, shipping with Research, Code, Design and Oncall agents plus 13 third-party presets through ACP, CLI and OpenAI-compatible APIs.
Anthropic launches Claude Code mods, enabling JavaScript or TypeScript plugins in version 2.1.287+ to add custom interfaces, pause dangerous commands such as rm -rf and git reset --hard, and replay file edits.
nanaismが、AI生成日本語を自然で高密度な文へ推敲するAgent Skill「yomiyasu」をMITライセンスで公開し、SVOCM復元など7原則と、AIらしい過剰な太字・箇条書き・絵文字を検査するPython製lintツールを提供する。
VEED launches OpenEdit, an Apache-2.0 open-source pipeline that lets Claude Code, Codex and Gemini CLI turn natural-language prompts into rendered MP4 edits without a GUI or timeline, using transcript-driven HTML compositions and segment-only re-rendering.
Anthropic now makes Claude for Government generally available in a FedRAMP High-authorized environment for U.S. federal and state agencies, adding early access to Claude Code CLI and Claude for Microsoft 365 with hard spending caps and agency-controlled conversation history.
Google unveils Gemini 4 Argon, a flagship AI model it claims leads rivals in coding, knowledge work and cybersecurity, with restricted Fairwind access for cyber defenders and pricing starting at $2 per million input tokens.
Google unveils Gemini 4 Argon, a cyber-defense model it says can autonomously find, validate and patch critical vulnerabilities, initially restricting access to trusted defenders in its Fairwind Program after Wiz uses it to uncover a healthcare-software flaw earlier frontier models missed.
Google unveils Gemini 4 Argon for vetted cybersecurity teams, claiming it leads GPT-6 Astra and Claude Opus 5.5 on 13 of 19 internal benchmarks and reaches 77.9% on DeepSWE coding.
Cognition slashes Devin pricing by up to 70% for Devin Review and launches a faster Cloud harness that batches and parallelizes tool calls, while Fusion averages $0.60 per FrontierCode 1.1 Extended task.
Baseten partners with OpenAI to bring open models natively to Codex and the Responses API, allowing enterprise customers to use existing OpenAI commitments while routing workloads across more than 90 active-active clusters and giving coding agents isolated Blaxel sandboxes.