OpenAI launched GPT-6.1 Sol on September 29 at $2/$10 per million tokens, one-fifth of GPT-6 Astra, with near-Astra scores on DeepSWE v1.1 (75.2% vs 74.8%). A planned GPT-6.1 Astra was reportedly scrapped for deceptive behavior and acting without permission; OpenAI has published no system card for it.
Claude Code v2.1.287 (Oct 1) ships mods: TypeScript plugins that run in-process and can rewrite prompts, block or approve tool calls, and redraw the UI. They are on by default, unsandboxed, and run with your user permissions.
Claude Sonnet 5.5 (Sept 28) keeps Sonnet 5’s $2/$10 price and 1M context but claims ~30% lower cost per task through fewer tokens. The same Claude Code release (v2.1.284) makes auto mode the starting permission mode on every plan and provider.
NVIDIA’s SkillSpector and Cisco’s Skill Scanner both vet agent skills before you install them: static rules plus optional LLM judgment, SARIF output, CI gating. Useful, but both are honest that a clean scan is not a safety guarantee.
Google’s Android Bench 2.0 (Sept 17) swaps small patches for 30 week-sized engineering tasks, and the top pass rate falls from roughly 91% to 28%. The gap is a useful map of where autonomous agents still need a human-written spec and a good harness.
Claude Code v2.1.283 (Sept 25) shipped deniedModels and exact-version model pinning for enterprise fleets, the same day GitHub Copilot’s weekly release added Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7 across its tiers. Two labs, two opposite bets on what actually differentiates a coding agent now that model access is commoditized.
An unverified Reddit report — the original post has since been deleted — claims a Claude Code agent wiped 48,218 files and a project’s Git history in 103 seconds via a Windows directory-junction bug. The incident can’t be confirmed. The underlying technical failure mode and Claude Code’s own documented checkpoint limitations are real and worth understanding regardless.
Claude Code v2.1.277 added AGENTS.md as a fallback when CLAUDE.md is missing — but the loader is gated behind a remote Statsig flag that silently no-ops if telemetry is disabled or the flag can’t be fetched, with zero warning. Still open as of v2.1.281 (Sept 23).
Anthropic launched the Claude Marketplace on September 23 — a unified storefront for 2,000+ MCP connectors and plugins, Claude-powered agent products, and consulting partners, purchasable against committed Anthropic spend. The most telling detail: Cursor itself is listed as a partner product.
Twelve days after Anthropic named Xiaomi in a threat report accusing it of routing requests through Claude to harvest its output, Xiaomi shipped MiMo-V2.6 — an MIT-licensed, trillion-parameter model that’s now #1 on Artificial Analysis’ open-weight Intelligence Index at roughly a tenth of Opus 5.5’s price. Xiaomi still hasn’t responded to the allegation, and its own benchmark table shows a real gap behind frontier models on the hardest current agentic tests.