Benchmarks
326 sources · 1,256 models × 2,956 benchmarks · 154,484 scoresCo-Design Bench's graded completion over the same 273 tasks: after the verifier's refutation pass each hidden acceptance criterion scores 1 for PASS, 0.5 for PARTIAL and 0 for FAIL, averaged within a task and then over tasks with equal weight. It is never below strict completion, which gives PARTIAL no credit.
547th strongest of 779 fitted benchmarks
Signal-to-noise +3.91 for the Co-Design Bench family
Another benchmark represents this family in the fit.
| # | Model | Setting | Score | Source |
|---|---|---|---|---|
| 1 | GPT-6 Astra | not reported · Claude Code | 75.3 | |
| 2 | Claude Opus 5 | not reported · Claude Code | 71.9 | |
| 3 | Qwen3.8 Max (0902) | not reported · Claude Code | 71.6 | |
| 4 | Qwen3.8 27B | not reported · Claude Code | 70.9 | |
| 5 | Kimi K3 | not reported · Claude Code | 69.5 | |
| 6 | GPT-5.6 Sol | not reported · Claude Code | 67.9 | |
| 7 | Opus 4.8 | not reported · Claude Code | 64.9 | |
| 8 | GLM-5.3 Flash | not reported · Claude Code | 63.9 | |
| 9 | GLM-5.2 | not reported · Claude Code | 63.7 | |
| 10 | Grok 4.6 | not reported · Claude Code | 58.0 |