Research Supervision on Scientist-Bench
2Number of StagesMLR-Copilot
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MLR-CopilotGPU=Yes, LLM=GPT-42026.03 | 2 | — | |
| AI Scientist v1GPU=Yes, LLM=Claude Sonnet 3.52026.03 | 3 | 15 | |
| Agent LabLLM=GPT-4o, GPU=Config.2026.03 | 3 | 2.33 | |
| Agent LabLLM=o1-preview, GPU=Config.2026.03 | 3 | 13.1 | |
| AI-ResearcherGPU=Yes, LLM=Gemini 2.5 Pro2026.03 | 3 | — | |
| AI-Supervisor (efficient)GPU=No, LLM=Qwen-72B2026.03 | 5 | 8 | |
| AI-Supervisor (frontier)GPU=No, LLM=GPT-4o / Claude2026.03 | 5 | 50 | |
| AI-Supervisor (local)GPU=Consumer, LLM=LLaMA / DeepSeek2026.03 | 5 | 0 |