Active Tool Exploration on Earth-Bench
39,080Tokens per Question (AP)RAG
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RAGModel=GPT-52026.05 | 39,080 | 37,280 | 7,394 | 7,155 | |
| RS-ClawModel=Qwen3-32b2026.05 | 70,759 | 97,325 | 5,951 | 6,309 | |
| RAGModel=Qwen3-32b2026.05 | 105,383 | 116,778 | 8,999 | 9,579 | |
| RS-ClawModel=GPT-52026.05 | 107,428 | 100,302 | 7,945 | 7,708 | |
| RAGModel=DeepSeek-V3.12026.05 | 111,566 | 97,659 | 9,147 | 9,143 | |
| FlatModel=GPT-52026.05 | 134,076 | 129,499 | 21,833 | 21,754 | |
| RS-ClawModel=DeepSeek-V3.12026.05 | 158,433 | 159,347 | 8,116 | 8,199 | |
| FlatModel=DeepSeek-V3.12026.05 | 352,398 | 289,428 | 27,587 | 27,490 | |
| FlatModel=Qwen3-32b2026.05 | 502,119 | 517,445 | 30,612 | 31,756 |