Deep Information Search and Synthesis on xbench DeepSearch
77.8ScoreOpenAI-GPT-5-high
Evaluation Results
| Method | Links | |
|---|---|---|
| OpenAI-GPT-5-highModel Access Type=Closed-Source2026.02 | 77.8 | |
| Tongyi-DeepResearch 30BModel Access Type=Open-Source2026.02 | 75 | |
| Tongyi DeepResearch# Samples=?, # OS Samples=0, Training=CPT + SFT + RL, Academic=false2026.03 | 75 | |
| OpenSeeker-v1-30B-SFT# Samples=11.7 k, # OS Samples=11.7 k, Training=SFT, Academic=true2026.03 | 74 | |
| WebSailor-V2-30B-A3BModel Access Type=Open-Source2026.02 | 73.7 | |
| WebSailor-V2-30B# Samples=?, # OS Samples=0, Training=SFT + RL, Academic=false2026.03 | 73.7 | |
| Minimax-M2Model Access Type=Closed-Source2026.02 | 72 | |
| WebLeaper-30B-A3BModel Access Type=Open-Source2026.02 | 72 | |
| DeepSeek-V3.1-671B# Samples=?, # OS Samples=0, Training=?, Academic=false2026.03 | 71.2 | |
| DeepSeek-V3.2Model Access Type=Closed-Source2026.02 | 71 | |
| GLM-4.6Model Access Type=Closed-Source2026.02 | 70 | |
| AgentCPM-Explore-4BModel Access Type=Open-Source2026.02 | 70 | |
| Merged-Model-4BModel Access Type=Open-Source2026.02 | 69.9 | |
| Kimi-ResearcherModel Access Type=Closed-Source2026.02 | 69 | |
| Claude-4.5-SonnetModel Access Type=Closed-Source2026.02 | 66 | |
| WebLeaper-30B# Samples=15 k, # OS Samples=0, Training=SFT, Academic=false2026.03 | 66 | |
| WebSailor-V2-30B# Samples=?, # OS Samples=0, Training=SFT, Academic=false2026.03 | 61.7 | |
| MiroThinker 8BModel Access Type=Open-Source2026.02 | 60.6 | |
| WebSailor-32B# Samples=?, # OS Samples=0, Training=SFT + RL, Academic=false2026.03 | 53.3 | |
| DeepDive-32B# Samples=4.1 k, # OS Samples=4.1 k, Training=SFT + RL, Academic=false2026.03 | 51.8 | |
| ASearcher-QWQ-32B v2Model Access Type=Open-Source2026.02 | 51.1 | |
| WebDancer-QwQ-32BModel Access Type=Open-Source2026.02 | 38.3 |