List your company

Multimodal Agent Execution on VisualClawArena (3,106-round core)

52.03Early AccuracyVisualClaw (Cat.)
39.726842.920946.11549.3091Jun 15, 2026

Evaluation results

MethodLinks
VisualClaw (Cat.)Backend=Claude Code, Setting=retrieved failures concatenated with skills2026.0652.0350.9153.557.7952.16
VisualClaw (Guide)Backend=Claude Code, Setting=memory-guided skill evolution2026.0650.8749.9851.456.5450.77
VisualClaw (Guide)Backend=Codex, Setting=memory-guided skill evolution2026.0650.6654.6656.8859.553.89
VisualClaw (Cat.)Backend=Codex, Setting=retrieved failures concatenated with skills2026.0649.6956.1957.3859.8854.27
VisualClaw w/o FullEvoBackend=Claude Code2026.0648.848.5149.6855.3449
VisualClaw w/o FullEvoBackend=Codex2026.0648.152.9353.4757.9551.35
Uniform-8Backend=Codex2026.0646.7450.4953.956.0850.25
Uniform-8 w/o FullEvoBackend=Codex2026.0644.3549.0452.5855.1548.51
Uniform-8Backend=Claude Code2026.0640.2544.4947.5249.143.99
Uniform-8 w/o FullEvoBackend=Claude Code2026.0640.241.5745.8747.4942.37