Agent Task on GTA
24.89Success RatePAIR
Evaluation Results
| Method | Links | |
|---|---|---|
| PAIR2026.05 | 24.89 | |
| LLM-as-a-judge2026.05 | 23.87 | |
| Multi-Attn2026.05 | 22.18 | |
| Attention2026.05 | 21.54 | |
| IGPO2026.05 | 20.05 | |
| Head-entropy2026.05 | 19.89 | |
| Tree-GRPO2026.05 | 19.34 | |
| AT2PO2026.05 | 18.65 | |
| TIPS2026.05 | 18.56 | |
| Outcome2026.05 | 18.51 | |
| Lookback-ratio2026.05 | 18.12 | |
| Mean-pooled2026.05 | 17.98 | |
| Multi-layer2026.05 | 17.89 | |
| CoE-C2026.05 | 17.21 | |
| Hidden+Attn2026.05 | 14.32 | |
| Last-token2026.05 | 13.72 |