Benchmarks
326 sources · 1,256 models × 2,956 benchmarks · 154,484 scoresStrict completion on Co-Design Bench's 100 Multimodal Editing tasks: localized visual, structural and content edits to a runnable page, given an annotated target region and, for 80 tasks, an image, video, audio or document attachment. Its 553 hidden criteria, about 5.5 per task, also check that unrelated content is preserved when the request asks for it.
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 | 75.4 | |
| 2 | Claude Opus 5 | not reported · Claude Code | 69.5 | |
| 3 | Qwen3.8 Max (0902) | not reported · Claude Code | 66.7 | |
| 4 | Kimi K3 | not reported · Claude Code | 66.2 | |
| 5 | Qwen3.8 27B | not reported · Claude Code | 64.2 | |
| 6 | Opus 4.8 | not reported · Claude Code | 59.3 | |
| 7 | GPT-5.6 Sol | not reported · Claude Code | 58.3 | |
| 8 | GLM-5.2 | not reported · Claude Code | 57.2 | |
| 9 | GLM-5.3 Flash | not reported · Claude Code | 53.8 | |
| 10 | Grok 4.6 | not reported · Claude Code | 46.7 |