OpenClaw adds a decisionModel role for typed judgments
OpenClaw introduces a decisionModel role for typed choice, score, and boolean checks, callable from plugins without another full chat completion.
OpenClaw introduces a decisionModel role for typed choice, score, and boolean checks, callable from plugins without another full chat completion.
Anthropic shipped Claude Opus 5.5 for long-running coding agents, claiming Fable-level work at about 40% less than Opus 5, with same-day AWS access.
OpenAI is working with an unpaid IAS-hosted math advisory group on how AI-generated results are reviewed and released.
DigitalOcean’s Managed Agents public preview runs Claude Code, Codex, and LangGraph in idle-pausing microVMs with a 16,000+ tool MCP gateway.
Isaac ROS 5.0, announced at ROSCon, adds agent-ready robotics skills, ROS 2 Lyrical, and Jetson support from Orin Nano to Thor.
OpenAI launches GPT-6 Sol and Luna at 50% below GPT-5.6 API promo prices; independent benches show Sol up and Luna down on coding agents.
OpenAI’s Sept 21 post urges US-led global technical standards for frontier AI and recursive self-improvement, without licenses or pre-release model approval.
Xiaomi open-sources MiMo-V2.6 Pro and Flash: omni MoE models with 1M context, aimed at long-horizon coding and agent workloads.
Zuckerberg announced a Shopify team-up to make shopping and checkout easier in Muse. Shopify plans the integration with Shop Pay; operational details remain unknown.
SpaceXAI’s Grok 4.7 lands in Cursor, Grok Build, and the API at Grok 4.6 prices, with longer-task training and a native Grok Bot harness.
Amazon cut off Meta’s Muse shopping agent, citing unauthorized access, stored logins, and Conditions of Use after asking Meta to opt Amazon out.
UC Berkeley’s HarnessTax finds Claude Code, Codex CLI, and Pi change coding-agent success little while token cost can rise about 5×.
13–19 Sep 2026
SpaceXAI released Grok Voice Transcribe 2.0 for batch and streaming STT at $0.10/$0.20 per hour; docs already list 2.0 as the Speech-to-Text API default.
MIT and Sakana’s SIFT paper uses LLM judges and tree search so coding agents self-modify with far less full-benchmark evaluation.
Google Labs turns CC into a shared household agent with its own Google account for up to six U.S. adults, plus a daily brief and waitlist.
Claude Code 2.1.278 defaults auto-mode checks to a server-side classifier for API, Enterprise, and cloud paths, with no classifier overhead when that path works.
Playwright says pick MCP or CLI; docs spell out best-for, token cost, headed vs headless, and how to install.
Claude Code 2.1.277 reads AGENTS.md when CLAUDE.md is missing, with a /config toggle, plus gateway proxy headers and hang-exit fixes.
Anthropic turned Claude Code Projects into one conversation that coordinates parallel cloud threads and keeps working after you close the laptop.
Google opened Home MCP to Claude, OpenClaw, and other agents for U.S. Premium Advanced homes; door unlocks are blocked and Google warns the wrong agent can still misbehave.
Claude leads 26% of AI R&D as of August; ~30k agents are monitored; a July 13–20 week put about 6% of R&D compute on safety.
Claude Code 2.1.276 restores proxy and gateway traffic after 2.1.275 400s on Input tag advisor_20260301.
Claude Code 2.1.275 adds send-now to flush queued mid-turn messages and syncs claude.ai skills/plugins; it also fixes a terminal API 400 behind rewriting gateways.
NVIDIA says TensorRT Edge-LLM finished MLPerf Edge Agentic 6.4× faster than llama.cpp on Jetson AGX Thor, at 52.33 tok/s on Qwen3.6-27B.