Comparison on Held-out games (test)
96.8AccuracyBG-Critic
Evaluation results
| Method | Links | |
|---|---|---|
| BG-CriticStage 2 multi-task training=Included2026.06 | 96.8 | |
| GPT-5.42026.06 | 92.1 | |
| Qwen3.5-27BBackbone Status=untuned2026.06 | 90.2 | |
| Gemini-3.1-Flash2026.06 | 88.7 | |
| Qwen3.5-397B2026.06 | 85.5 | |
| BG-CriticStage 2 multi-task training=Excluded2026.06 | 73.1 |