Agentic Performance Aggregate on Combined LCtx, ALFWorld, ScienceWorld, Web/Tool
78.9Mean Task SuccessCVT-RL
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CVT-RLVerifier=yes, Constraints=yes, CF rollouts=semantic, Belief=yes, Compute=3.92026.06 | 78.9 | 82.8 | 3.9 | 0.175 | |
| Information-matched CF processVerifier=yes, Constraints=yes, CF rollouts=semantic, Belief=yes, Compute=3.92026.06 | 75.4 | 78.9 | 7.2 | 0.224 | |
| Compute-matched non-causalVerifier=yes, Constraints=yes, CF rollouts=random extra, Belief=yes, Compute=3.82026.06 | 71.8 | 74.6 | 11.7 | 0.271 | |
| Constrained process RLVerifier=yes, Constraints=yes, CF rollouts=no, Belief=yes, Compute=1.52026.06 | 69.5 | 73.8 | 12.6 | 0.286 | |
| RLVMRVerifier=yes, Constraints=no, CF rollouts=no, Belief=no, Compute=1.32026.06 | 68.7 | 72.9 | 13.5 | 0.305 | |
| LongRLVRVerifier=yes, Constraints=no, CF rollouts=no, Belief=no, Compute=1.22026.06 | 67.1 | 70.8 | 14.8 | 0.339 | |
| Q-RAGVerifier=yes, Constraints=no, CF rollouts=no, Belief=no, Compute=1.42026.06 | 66.8 | 73.5 | 14.1 | 0.332 | |
| TROLL-styleVerifier=yes, Constraints=no, CF rollouts=no, Belief=no, Compute=1.12026.06 | 63.5 | 66.4 | 15.9 | 0.351 | |
| PPO-RLVRVerifier=yes, Constraints=no, CF rollouts=no, Belief=no, Compute=1.12026.06 | 58.4 | 63.1 | 17.4 | 0.384 | |
| SFTVerifier=no, Constraints=no, CF rollouts=no, Belief=no, Compute=1.02026.06 | 47.2 | 55.8 | 18.9 | 0.412 |