Agentic Web Interaction on BrowseComp-ZH (test)
61.3Pass@1GPT-5
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-5Agent Category=Proprietary Agents2026.04 | 61.3 | — | |
| DeepSeek-V3.2Agent Category=Proprietary Agents2026.04 | 53.6 | — | |
| DeepSeek-V3.1Agent Category=Proprietary Agents2026.04 | 49.5 | — | |
| GLM-4.6Agent Category=Proprietary Agents2026.04 | 42.2 | — | |
| LThinker++Agent Category=Our Agents, Protocol=Behavioral Pruning constraints2026.04 | 36.9 | 57.1 | |
| Vanilla-AgentAgent Category=Our Agents, Protocol=+ SFT (vanilla)2026.04 | 31.5 | 47.8 | |
| Claude-4-SonnetAgent Category=Proprietary Agents2026.04 | 29.1 | — | |
| Kimi-K2-InstructAgent Category=Proprietary Agents2026.04 | 28.8 | — | |
| Qwen3-235B-A22B-InstructAgent Category=Proprietary Agents2026.04 | 21.8 | — | |
| Qwen3-30B-A3B-ThinkingAgent Category=Our Agents, Protocol=Base2026.04 | 10 | 17.3 |