GPT 6.2 Astra and Sol Take 4th and 5th on sudo L7 Coding Benchmark, Still Behind Latest Claude
echen · x · 2026-11-18
GPT 6.2 Astra (38.3%) and GPT 6.2 Sol (36.1%) now rank 4th and 5th on sudo L7, a benchmark for staff-level coding, ahead of earlier GPT releases but still well behind the latest Claude models.
Full leaderboard: Claude Fable 5.2 leads at 52.2%, followed by Claude Sonnet 5.5 (50.6%) and Claude Opus 5.5 (46.1%), sweeping the top three. Further down, GPT 6.1 Sol scores 33.9%, Claude Fable 5.1 32.2%, GPT 6 Astra 31.1%, MiniMax M3.1 Flash 18.9%, and Qwen 3.8 Max 16.7%.
More from Models
- GPT-6.2 Astra not obviously better than GPT-6 at Blood on the Clocktower eval — SamuelAlbanie · 2026-11-18
- Gemini meeting transcripts auto-mask swearwords, HF researcher plans to stress test it — _lewtun · 2026-11-18
- Hands-on with GPT-6.2: Ignore the hate, it's a cheap computer-use workhorse — nickbaumann_ · 2026-11-18
- Claude leads Arena agent leaderboard with nearly 2x praise-to-complaint ratio — thursdai_pod · 2026-11-18
- GPT-6 Cyber uncovers critical Java bug affecting ~90% of Fortune 500, Calif says — gdb · 2026-11-18
- GPT-6.2 Astra panned by top reviewers: 'essentially unusable,' falls short of Opus 5.5 and Fable — deedydas · 2026-11-18