Multimodal Agent Execution on VisualClawArena (3,106-round core)
52.03Early AccuracyVisualClaw (Cat.)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VisualClaw (Cat.)Backend=Claude Code, Setting=retrieved failures concatenated with skills2026.06 | 52.03 | 50.91 | 53.5 | 57.79 | 52.16 | |
| VisualClaw (Guide)Backend=Claude Code, Setting=memory-guided skill evolution2026.06 | 50.87 | 49.98 | 51.4 | 56.54 | 50.77 | |
| VisualClaw (Guide)Backend=Codex, Setting=memory-guided skill evolution2026.06 | 50.66 | 54.66 | 56.88 | 59.5 | 53.89 | |
| VisualClaw (Cat.)Backend=Codex, Setting=retrieved failures concatenated with skills2026.06 | 49.69 | 56.19 | 57.38 | 59.88 | 54.27 | |
| VisualClaw w/o FullEvoBackend=Claude Code2026.06 | 48.8 | 48.51 | 49.68 | 55.34 | 49 | |
| VisualClaw w/o FullEvoBackend=Codex2026.06 | 48.1 | 52.93 | 53.47 | 57.95 | 51.35 | |
| Uniform-8Backend=Codex2026.06 | 46.74 | 50.49 | 53.9 | 56.08 | 50.25 | |
| Uniform-8 w/o FullEvoBackend=Codex2026.06 | 44.35 | 49.04 | 52.58 | 55.15 | 48.51 | |
| Uniform-8Backend=Claude Code2026.06 | 40.25 | 44.49 | 47.52 | 49.1 | 43.99 | |
| Uniform-8 w/o FullEvoBackend=Claude Code2026.06 | 40.2 | 41.57 | 45.87 | 47.49 | 42.37 |