Web/Tool Tasks on AgentBench 700 (test)
79.7Success RateCVT-RL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CVT-RLDescription=per-seed CVT-RL success, Split (train/dev/test)=4200/520/700, Individual Seeds=78.4/80.2/79.9/78.7/81.32026.06 | 79.7 | 3.8 | |
| IMDescription=information-matched counterfactual-process RL, Split (train/dev/test)=4200/520/7002026.06 | 75.9 | — | |
| CMDescription=compute-matched non-causal RL, Split (train/dev/test)=4200/520/7002026.06 | 72.8 | — |