vibehacker
News
METR ·

METR: Anthropic model capabilities improving ~2x faster, mostly via AI-assisted R&D

In a Nov 5 high-access assessment, METR estimates Anthropic is improving capabilities on well-specified tasks about twice as fast as March 2024–October 2025 (AECI ~1.9x), with most of today’s acceleration attributed to AI-assisted R&D rather than training-compute scaling. Anthropic redacted much of the evidence; METR saw no clear near-term bottlenecks that would stop further acceleration.

More news

View all

Qwen Code 0.26.9 adds Skill ZIP installs and bulk Skill activation

Released Nov 6, Qwen Code 0.26.9 lets you install large Skill ZIPs via the daemon API, TypeScript SDK, or Web Shell, and toggle 1–100 Skills in one PUT /skills/activation call. It also turns managed agent automation on by default, adds image input for Llama 4, Pixtral, and Mistral Small models, and reports working/blocked/done status to terminals via OSC 7501…

Apiiro: Claude Code "don't ask again" left secrets in 1,122 public GitHub copies

Apiiro found 1,122 public .claude/settings.local.json copies holding at least one secret (1,379 distinct, including 379 database passwords) after "Yes, and don't ask again" saves the full approved command—credentials included—in plain text. Claude Code only adds the file to the user's global git excludes, not the project's .gitignore , so already committed copies and forks stay exposed; rotate first, then clean up…

The Hacker News

Spotted something we missed? Start a thread.