Benchmarks · reasoning
FrontierChallenge
Uses the first-party pass rate across 97 scientific workflow tasks. Score: pass rate. Collected by Jake Boggs' benchmark dashboard.
11 models · 11 results · data as of 2026-10-29 · apodexai.github.io ↗
Leaderboard
GPT-5.6 Sol max
All results
11 rows| # | Model | Effort | FrontierChallenge |
|---|---|---|---|
| 1 | GPT-5.6 Sol | max | 20.6% |
| 2 | Grok 4.6 | max | 20.6% |
| 3 | Opus 5 | max | 17.5% |
| 4 | Kimi K3 | max | 17.5% |
| 5 | GPT-5.6 Terra | max | 15.5% |
| 6 | Qwen 3.8 Max | max | 15.5% |
| 7 | DeepSeek V4 Pro 0813 | max | 13.4% |
| 8 | DeepSeek V4 Flash 0731 | max | 12.4% |
| 9 | Gemini 3.7 Flash | max | 10.3% |
| 10 | Qwen 3.5 397B-A17B | max | 4.1% |
| 11 | GLM 5.2 | max | 3.1% |