Data Science Agent tasks on xBench-DS
73Pass@1MIRO-30B + MIROFLOW
Evaluation Results
| Method | Links | |
|---|---|---|
| MIRO-30B + MIROFLOWCategory=OPEN-SOURCE2026.03 | 73 | |
| MIRO-30B + WEDASCategory=OPEN-SOURCE2026.03 | 72 | |
| WEBSAILOR-72BCategory=OPEN-SOURCE2026.03 | 55 | |
| WEBSAILOR-32BCategory=OPEN-SOURCE2026.03 | 53.3 | |
| GROK-DEEPRESEARCHCategory=CLOSED-SOURCE2026.03 | 52 | |
| DOUBAO-DEEPTHINKCategory=CLOSED-SOURCE2026.03 | 50 | |
| GPT-5-MINI + WEDASCategory=OPEN-SOURCE2026.03 | 44 | |
| ASEARCHER-WEB-32BCategory=OPEN-SOURCE2026.03 | 42.1 | |
| WEBDANCER-QWQ-32BCategory=OPEN-SOURCE2026.03 | 38.3 | |
| GPT-5-MINI + MIROFLOWCategory=OPEN-SOURCE2026.03 | 38 | |
| SEARCH-O1-32BCategory=OPEN-SOURCE2026.03 | 25 | |
| WEBTHINKER-32B-RLCategory=OPEN-SOURCE2026.03 | 24 | |
| O4-MINICategory=DIRECT INFERENCE2026.03 | 22.3 | |
| GPT-4OCategory=DIRECT INFERENCE2026.03 | 18 | |
| GPT-4.1Category=DIRECT INFERENCE2026.03 | 17 | |
| QWEN-2.5-72BCategory=DIRECT INFERENCE2026.03 | 12.7 | |
| QWQ-32BCategory=DIRECT INFERENCE2026.03 | 10.7 | |
| QWEN-2.5-32BCategory=DIRECT INFERENCE2026.03 | 8.7 | |
| TodoEvolve + SmolagentsModel Family=GPT-5-Mini, pass@k protocol=pass@12026.02 | 0.75 | |
| TodoEvolve + SmolagentsModel Family=DeepSeek V3.2, pass@k protocol=pass@12026.02 | 0.74 | |
| TodoEvolve + SmolagentsModel Family=Kimi K2, pass@k protocol=pass@12026.02 | 0.71 | |
| Flash-SearcherModel Family=GPT-5-mini, pass@k protocol=pass@12026.02 | 0.69 | |
| Agent KBModel Family=GPT-4.1, pass@k protocol=pass@32026.02 | 0.68 | |
| Flash-SearcherModel Family=DeepSeek V3.2, pass@k protocol=pass@12026.02 | 0.68 | |
| Flash-SearcherModel Family=Kimi K2, pass@k protocol=pass@12026.02 | 0.66 | |
| Agent KBModel Family=GPT-4.1, pass@k protocol=pass@22026.02 | 0.58 | |
| Cognitive Kernel-ProModel Family=Claude-3.7 etc., pass@k protocol=pass@12026.02 | 0.56 | |
| OWL WorkforceModel Family=GPT-4o+o3-mini, pass@k protocol=pass@32026.02 | 0.55 | |
| SmolagentsModel Family=GPT-5-mini, pass@k protocol=pass@12026.02 | 0.51 | |
| Agent KBModel Family=GPT-4.1, pass@k protocol=pass@12026.02 | 0.48 | |
| OAgentsModel Family=Claude-3.7 etc., pass@k protocol=pass@12026.02 | 0.47 |