Mean Reward on ScienceWorld
0.319Mean RewardTrunc
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TruncBackbone=gpt-5-mini, Inference Protocol=Inference-free2026.05 | 0.319 | 107 | |
| NO-COMPBackbone=gpt-5-mini, Inference Protocol=Inference-free2026.05 | 0.298 | — | |
| TruncBackbone=qwen3.5-flash, Inference Protocol=Inference-free2026.05 | 0.285 | 124 | |
| NO-COMPBackbone=qwen3.5-flash, Inference Protocol=Inference-free2026.05 | 0.23 | — | |
| LongLinguaBackbone=gpt-5-mini, Inference Protocol=Inference-free2026.05 | 0.229 | 99 | |
| AGORABackbone=gpt-5-mini, Inference Protocol=Inference-free2026.05 | 0.217 | 73 | |
| ObsMaskBackbone=gpt-5-mini, Inference Protocol=Inference-free2026.05 | 0.215 | 93 | |
| AGORABackbone=qwen3.5-flash, Inference Protocol=Inference-free2026.05 | 0.21 | 91 | |
| HiAgentBackbone=gpt-5-mini, Inference Protocol=LLM-call per step2026.05 | 0.206 | 89 | |
| LongLinguaBackbone=qwen3.5-flash, Inference Protocol=Inference-free2026.05 | 0.201 | 103 | |
| Floor-K2Backbone=gpt-5-mini, Inference Protocol=Inference-free2026.05 | 0.188 | 63 | |
| NO-COMPBackbone=gpt-4o-mini, Inference Protocol=Inference-free2026.05 | 0.173 | — | |
| ACONBackbone=gpt-5-mini, Inference Protocol=LLM-call per step2026.05 | 0.172 | 74 | |
| AgentDietBackbone=gpt-5-mini, Inference Protocol=LLM-call per step2026.05 | 0.168 | 72 | |
| HiAgentBackbone=qwen3.5-flash, Inference Protocol=LLM-call per step2026.05 | 0.164 | 84 | |
| Floor-K2Backbone=gpt-4o-mini, Inference Protocol=Inference-free2026.05 | 0.164 | 95 | |
| ObsMaskBackbone=qwen3.5-flash, Inference Protocol=Inference-free2026.05 | 0.161 | 82 | |
| TruncBackbone=gpt-4o-mini, Inference Protocol=Inference-free2026.05 | 0.156 | 90 | |
| AGORABackbone=gpt-4o-mini, Inference Protocol=Inference-free2026.05 | 0.155 | 89 | |
| ObsMaskBackbone=gpt-4o-mini, Inference Protocol=Inference-free2026.05 | 0.154 | 117 | |
| Floor-K2Backbone=qwen3.5-flash, Inference Protocol=Inference-free2026.05 | 0.147 | 64 | |
| AgentDietBackbone=qwen3.5-flash, Inference Protocol=LLM-call per step2026.05 | 0.147 | 75 | |
| ACONBackbone=qwen3.5-flash, Inference Protocol=LLM-call per step2026.05 | 0.146 | 75 | |
| HiAgentBackbone=gpt-4o-mini, Inference Protocol=LLM-call per step2026.05 | 0.146 | 111 | |
| LongLinguaBackbone=gpt-4o-mini, Inference Protocol=Inference-free2026.05 | 0.144 | 110 | |
| ACONBackbone=gpt-4o-mini, Inference Protocol=LLM-call per step2026.05 | 0.132 | 100 | |
| AgentDietBackbone=gpt-4o-mini, Inference Protocol=LLM-call per step2026.05 | 0.129 | 98 | |
| SelCtxBackbone=qwen3.5-flash, Inference Protocol=Inference-free2026.05 | 0.114 | 58 | |
| SelCtxBackbone=gpt-5-mini, Inference Protocol=Inference-free2026.05 | 0.114 | 49 | |
| SelCtxBackbone=gpt-4o-mini, Inference Protocol=Inference-free2026.05 | 0.091 | 69 |