Deep Research Instruction Following on DeepResearchIF
56.4Query-Level AccuracyGPT-5.2
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GPT-5.22026.06 | 56.4 | 89.2 | 71.6 | |
| Claude-4.6-Sonnet-Thinking2026.06 | 33.6 | 81.1 | 59.6 | |
| DeepSeek-V3.2Size=671B2026.06 | 25.8 | 76 | 54.8 | |
| S1-DeepResearchSize=32B, Train=SFT2026.06 | 25.2 | 74.3 | 54.2 | |
| Gemini-3.1-Pro-Preview2026.06 | 24.8 | 71.3 | 56.2 | |
| Doubao-2.0-Pro2026.06 | 23.4 | 77.7 | 55.4 | |
| Kimi-K2.5Size=1T2026.06 | 18.9 | 74.3 | 51.2 | |
| Qwen3.5-397BSize=397B2026.06 | 14.6 | 68.4 | 50.4 | |
| MiniMax-M2.7Size=230B2026.06 | 11 | 63.9 | 45.1 | |
| GLM-5Size=744B2026.06 | 10.8 | 62.7 | 47.9 | |
| Qwen3-235BSize=235B2026.06 | 7.4 | 54.9 | 43.6 | |
| Qwen3-32BSize=32B2026.06 | 4.1 | 28.9 | 40.6 |