Deep Research on GAIA (Pass@1, Accuracy)
51.46Pass@1THINKMERGE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| THINKMERGEmodel=WebSailor-32B, N=42025.12 | 51.46 | — | |
| THINKMERGEmodel=WebSailor-32B, N=82025.12 | 50.49 | — | |
| THINKMERGEmodel=WebSailor-32B, N=22025.12 | 48.55 | — | |
| WebSailor-32Bvariant=Baseline2025.12 | 46.64 | — | |
| THINKMERGEmodel=WebSailor-7B, N=42025.12 | 41.26 | — | |
| THINKMERGEmodel=WebSailor-7B, N=82025.12 | 36.89 | — | |
| WebSailor-7Bvariant=Baseline2025.12 | 35.52 | — | |
| THINKMERGEmodel=WebSailor-7B, N=22025.12 | 33.98 | — | |
| Youtu-LLM 2BThinking Mode=true, ReAct Framework=true, LLM-as-a-Judge=true2025.12 | 33.9 | 33.9 | |
| THINKMERGEmodel=WebSailor-3B, N=22025.12 | 33.49 | — | |
| WebSailor-3Bvariant=Baseline2025.12 | 32.22 | — | |
| Qwen3 4BThinking Mode=true, ReAct Framework=true, LLM-as-a-Judge=true2025.12 | 25.5 | 25.5 | |
| THINKMERGEmodel=WebSailor-3B, N=42025.12 | 15.04 | — | |
| SmolLM3 3BThinking Mode=true, ReAct Framework=true, LLM-as-a-Judge=true2025.12 | 11.7 | 11.7 | |
| Qwen3 1.7BThinking Mode=true, ReAct Framework=true, LLM-as-a-Judge=true2025.12 | 11.4 | 11.4 | |
| THINKMERGEmodel=WebSailor-3B, N=82025.12 | 5.34 | — |