Claude Code Opus 5.5 tracker: 81% 30-day pass rate, degradation checks paused
gleech · x · 2026-11-10
Marginlab's Claude Code Opus 5.5 tracker benchmarks daily in the Claude Code CLI on a curated SWE-Bench-Pro subset: 77% daily pass rate (50 cases), 76% over 7 days (350), 81% over 30 days (1,200). Degradation detection is paused while a new baseline is collected after the Sept 24 model switch. Commenters link the gap to this week's sycophantic 'You're absolutely right' Opus sessions.
More from Models
- Anthropic's new model formalized Fermat's Last Theorem in 30 hours and built a bootable OS in 8 days — kevinnbass · 2026-11-10
- Anthropic reportedly injects hidden ip_reminder via API, then blocks user probing it as 'reverse engineering' — OwariDa · 2026-11-10
- Anthropic system card reveals Claude reasoning about RL training, 79% eval awareness rate — gleech · 2026-11-10
- KVARK Sparkle 1.1 (27B, Apache 2.0): Refuses When Docs Lack the Answer, Uses 7-48% Fewer Tokens — dzulajla98 · 2026-11-10
- Should Google ship rumored Krypton immediately, or hoard it for internal use? — BlackHC · 2026-11-10
- Hugging Face open-sources 14.8B DNA model Carbon-X, second only to GPT-6.1 Sol on GenomeQA — LoubnaBenAllal1 · 2026-11-10