Agentic Web Interaction on xbench DeepSearch 2510 (test)
66Pass@1GPT-5
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-5Agent Category=Proprietary Agents2026.04 | 66 | — | |
| DeepSeek-V3.2Agent Category=Proprietary Agents2026.04 | 51 | — | |
| GLM-4.6Agent Category=Proprietary Agents2026.04 | 47 | — | |
| DeepSeek-V3.1Agent Category=Proprietary Agents2026.04 | 44 | — | |
| LThinker++Agent Category=Our Agents, Protocol=Behavioral Pruning constraints2026.04 | 44 | 60 | |
| Vanilla-AgentAgent Category=Our Agents, Protocol=+ SFT (vanilla)2026.04 | 38.3 | 53 | |
| Claude-4-SonnetAgent Category=Proprietary Agents2026.04 | 35 | — | |
| Kimi-K2-InstructAgent Category=Proprietary Agents2026.04 | 30 | — | |
| Qwen3-235B-A22B-InstructAgent Category=Proprietary Agents2026.04 | 27 | — | |
| Qwen3-30B-A3B-ThinkingAgent Category=Our Agents, Protocol=Base2026.04 | 8.7 | 16 |