Multi-turn tool use on GAIA
76.4Pass@1WebAnchor-30B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| WebAnchor-30BRL Algo=Anchor-GRPO2026.01 | 76.4 | 86.4 | |
| WebAnchor-30BRL Algo=GRPO2026.01 | 75.8 | 87.6 | |
| WebAnchor-30BRL Algo=First-step GRPO2026.01 | 74.8 | 87.4 | |
| OpenAI-o3†RL Algo=-2026.01 | 70.5 | — | |
| Claude-4-Sonnet†RL Algo=-2026.01 | 68.3 | — | |
| DeepSeek-V3.1†RL Algo=-2026.01 | 63.1 | — | |
| Kimi-K2†RL Algo=-2026.01 | 57.7 | — | |
| WebSailor-32BRL Algo=-2026.01 | 53.2 | — | |
| WebDancer-QwQRL Algo=-2026.01 | 51.5 | — | |
| WebThinker-RLRL Algo=-2026.01 | 48.5 | — | |
| WebAnchor-7BRL Algo=First-step GRPO2026.01 | 38.5 | 55.1 | |
| WebSailor-7BRL Algo=-2026.01 | 37.9 | — | |
| WebAnchor-7BRL Algo=Anchor-GRPO2026.01 | 37.8 | 51.5 | |
| WebAnchor-7BRL Algo=GRPO2026.01 | 33 | 44.7 | |
| WebAnchor-3BRL Algo=First-step GRPO2026.01 | 30.1 | 44.6 | |
| WebAnchor-3BRL Algo=Anchor-GRPO2026.01 | 28.3 | 43.5 | |
| WebAnchor-3BRL Algo=GRPO2026.01 | 27.2 | 43.6 | |
| R1-Searcher-7BRL Algo=-2026.01 | 20.4 | — |