Alignment with human judgment on game verification on VERIGAME
92.2Top-5 AccuracyGameGen-Verifier
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GameGen-VerifierBackend=Codex2026.05 | 92.2 | 91.2 | 100 | 95.4 | 443 | |
| GameGen-VerifierBackend=Claude Code (Opus)2026.05 | 90.2 | 89.2 | 100 | 94.3 | 513 | |
| GameGen-VerifierBackend=Claude Code (Kimi)2026.05 | 86.6 | 89.3 | 96.2 | 92.6 | 614 | |
| AaaV-CEBackend=Codex2026.05 | 58.8 | 96.8 | 48.4 | 64.5 | 7,356 | |
| AaaV-CEBackend=Claude Code (Opus)2026.05 | 57.6 | 93.5 | 47.5 | 63 | 8,245 | |
| AaaV-CEBackend=Claude Code (Kimi)2026.05 | 51 | 92.3 | 52.2 | 66.7 | 9,072 | |
| AaaV-DirectBackend=Claude Code (Kimi)2026.05 | 3.3 | 81.2 | 2.1 | 4 | 537 | |
| AaaV-DirectBackend=Claude Code (Opus)2026.05 | 1.3 | 71.4 | 1.1 | 2.2 | 552 | |
| AaaV-DirectBackend=Codex2026.05 | 1.1 | 66.7 | 0.9 | 1.8 | 387 |