MAS Test-Time Scaling on HotpotQA
57AccuracyOracle (UB)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Oracle (UB)Verification Level=Trajectory, N=82026.06 | 57 | — | |
| GPT-5.4-miniVerification Level=Orchestration, N=82026.06 | 46 | 7.75 | |
| GPT-5-miniVerification Level=Trajectory, N=82026.06 | 44 | 14.86 | |
| MAS-Orchestra (LB)Verification Level=-, N=82026.06 | 42.5 | — | |
| Maj. VoteVerification Level=Trajectory, N=82026.06 | 42.5 | — | |
| Orch-RMVerification Level=Orchestration, N=82026.06 | 42.5 | 7.32 | |
| GPT-4.1-miniVerification Level=Orchestration, N=82026.06 | 42 | 7.66 | |
| GPT-5.2Verification Level=Orchestration, N=82026.06 | 42 | 7.77 | |
| Skywork-V2-Qwen3-4BVerification Level=Orchestration, N=82026.06 | 42 | 7.32 | |
| Skywork-V2-Llama-3.2-3BVerification Level=Orchestration, N=82026.06 | 41.5 | 7.32 | |
| Skywork-V2-Qwen3-8BVerification Level=Orchestration, N=82026.06 | 41 | 7.32 | |
| GPT-5-miniVerification Level=Orchestration, N=82026.06 | 40.5 | 7.69 | |
| Skywork-LLaMA-3.1-8BVerification Level=Orchestration, N=82026.06 | 40.5 | 7.32 | |
| log PVerification Level=Orchestration, N=82026.06 | 39.5 | — |