Latest News#
- [2026-01-12]🚀🚀🚀 We have open-sourced AgentCPM-Explore, an agent foundation model with only 4B parameters, together with its entire training and inference infrastructure. AgentCPM-Explore has successfully entered 8 classic long-horizon agent benchmarks, including GAIA,HLE, and BrowserComp. AgentCPM-Explore achieves SOTA performance at the same parameter scale and demonstrates its accurate deep research capabilities, effectively breaking the performance bottleneck for on-device agents.
Overview#
Key highlights of AgentCPM-Explore include:
- The first full-parameter 4B agent model to rank on 8 long-horizon and complex agent benchmarks, including GAIA, HLE, and BrowserComp, in the on-device setting.
- Capable of over 100 rounds of continuous environment interaction, supporting multi-source information cross-validation, dynamic search strategy adjustment, and real-time verification of up-to-date information, enabling sustained deep exploration until task completion.
- Fully open-sourced end-to-end, including (1) AgentRL, a fully asynchronous reinforcement learning framework for agent training, (2) AgentDock, a unified management and scheduling platform for tool sandboxes, (3) AgentToLeaP, a one-click evaluation platform for agent tool-learning capabilities. These components collectively support community collaboration and custom extensibility.
We elaborate on the entire construction pipeline of AgentCPM-Explore on GitHub.
Experimental Results#
| Model | GAIA (text-only) | BrowseComp | BrowseComp (ZH) | HLE | Frames | WebWalker | Seal-0 | Xbench-DeepSearch
| Closed-Source Models
| Claude-4.5-sonnet | 71.2% | 19.6% | 40.8% | 24.5% | 85.0% | / | 53.4% | 66.0%
| Gemini Deep Research | / | / | / | 26.9% | / | / | / | /
| DeepSeek-V3.2 | 63.5% | 67.6% | 65.0% | 40.8% | 80.2% | / | 38.5% | 71.0%
| MiniMax-M2 | 75.7% | 44.0% | 48.5% | 31.8% | / | / | / | 72.0%
| OpenAI-GPT-5-high | 76.4% | 54.9% | 65.0% | 35.2% | / | / | 51.4% | 77.8%
| GLM-4.6 | 71.9% | 45.1% | 49.5% | 30.4% | / | / | / | 70.0%
| Kimi-Researcher | / | / | / | 26.9% | 78.8% | / | 36.0% | 69.0%
| Seed-1.8 | 87.4% | 67.6% | 81.3% | 40.9% | / | / | / | /
| Open-Source Models
| MiroThinker 8B | 66.4% | 31.1% | 40.2% | 21.5% | 80.6% | 60.6% | 40.4% | 60.6%
| Tongyi DeepResearch 30B | 70.9% | 43.4% | 46.7% | 32.9% | 90.6% | 72.2% | / | 75.0%
| ASearcher QWQ 32B v2 | 58.7% | / | / | / | 74.5% | / | / | 51.1%
| iterresearch-30B-A3B | 72.8% | 37.3% | 45.2% | 28.8% | 71.0% | / | 39.6% | /
| WebSailor-V2-30B-A3B (RL) | 74.1% | 35.3% | 44.1% | 30.6% | / | / | / | 73.7%
| WebLeaper-30B-A3B-RUC | 73.2% | 38.8% | / | / | / | / | 48.6% | 72.0%
| WebDancer (QWQ-32B) | 51.5% | 3.8% | 18.0% | / | / | 47.9% | / | 38.3%
| ⭐ AgentCPM-Explore 4B | 63.9% | 25.0% | 29.0% | 19.1% | 82.7% | 68.1% | 40.0% | 70.0%
