Open-ended Deep Research on Composite Suite (ScholarQA, HealthBench, ResearchQA, DeepResearchBench v2)
80.2ResearchQA AccuracySearchSwarm
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SearchSwarmSystem Category=Open-source models2026.06 | 80.2 | 79.2 | 52.8 | 44.4 | 64.2 | |
| OpenAI DeepResearchSystem Category=Closed-source systems2026.06 | 79.2 | 79.6 | 53.8 | 46.9 | 64.9 | |
| Dr.TuluSystem Category=Open-source models2026.06 | 75.7 | 88.3 | 52.8 | 45.4 | 65.6 | |
| Perplexity DeepResearchSystem Category=Closed-source systems2026.06 | 75.3 | 67.3 | — | 42.3 | — | |
| Gemini-3.1-Pro + searchSystem Category=Closed-source systems2026.06 | 74.5 | — | 47.5 | 44.4 | — | |
| WebThinker-32B-DPOSystem Category=Open-source models2026.06 | 74.2 | 46.7 | 39.4 | 40.6 | 50.2 | |
| Tongyi DeepResearchSystem Category=Open-source models2026.06 | 66.7 | 46.5 | 46.2 | 40.6 | 50 | |
| QwQ-32BSystem Category=Open-source models2026.06 | 60.9 | 41.9 | 24.5 | 40.3 | 41.9 | |
| Qwen3-8BSystem Category=Open-source models2026.06 | 56.1 | 40.4 | 16.5 | 33.3 | 36.6 |