Agent Capability Evaluation on SEAL 0
61.3Average Score (@8)MiroThinker-H1
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MiroThinker-H1Agent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 61.3 | — | |
| Kimi-K2.5Agent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 57.4 | — | |
| Claude-4.5-SonnetType=Foundation Models with Tools2025.11 | 53.4 | — | |
| MiroThinker-1.7Agent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 53 | — | |
| OpenAI-GPT-5Agent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 51.4 | — | |
| OpenAI-GPT-5-highType=Foundation Models with Tools2025.11 | 51.4 | — | |
| MiroThinker-v1.0-72BParameters=72B2025.11 | 51 | — | |
| DeepSeek-V3.2Agent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 49.5 | — | |
| Seed-2.0-ProAgent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 49.5 | — | |
| MiroThinker-1.7-miniAgent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 48.2 | — | |
| Claude-4.5-OpusAgent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 47.7 | — | |
| Qwen3.5-397BAgent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 46.9 | — | |
| MiroThinker-v1.0-30BParameters=30B2025.11 | 46.8 | — | |
| Gemini-3.0-ProAgent Style=ReAct, Max interaction turns=200, Max episode retries=5, Context management retention=52026.03 | 45.5 | — | |
| MiroThinker-v1.0-8BParameters=8B2025.11 | 40.4 | — | |
| DeepSeek-V3.2Type=Foundation Models with Tools2025.11 | 38.5 | — | |
| Kimi-ResearcherType=Research Agents2025.11 | 36 | — | |
| Kimi-K2-0905Type=Foundation Models with Tools2025.11 | 25.2 | — | |
| OpenAI-o3Type=Foundation Models with Tools2025.11 | 17.1 | — | |
| AgentCPM-Explore-4BModel Access Type=Open-Source2026.02 | — | 40.5 | |
| Claude-4.5-SonnetModel Access Type=Closed-Source2026.02 | — | 53.4 | |
| DeepSeek-V3.2Model Access Type=Closed-Source2026.02 | — | 38.5 | |
| IterResearch-30B-A3BModel Access Type=Open-Source2026.02 | — | 39.6 | |
| Kimi-ResearcherModel Access Type=Closed-Source2026.02 | — | 36 | |
| Merged-Model-4BModel Access Type=Open-Source2026.02 | — | 35.9 | |
| MiroThinker 8BModel Access Type=Open-Source2026.02 | — | 40.4 | |
| OpenAI-GPT-5-highModel Access Type=Closed-Source2026.02 | — | 51.4 | |
| WebLeaper-30B-A3BModel Access Type=Open-Source2026.02 | — | 48.6 |