JiYuan LüDòng, a startup founded by former Huawei Noah's Ark Lab director Wang Yunhe, has released its first Agent-Native model, NeoHorse. This model, developed with support from Wuxin Qiong and research contributions from Tsinghua and Peking Universities, focuses on enhancing an agent's ability to utilize tools, interpret environmental feedback, and correct errors. NeoHorse-1, available in 4B and 9B versions, has shown significant improvements after Agentic Post-Training, with the 4B version now matching or exceeding its 9B counterpart on various benchmarks, particularly in tasks with clear workflows and verifiable outcomes. AI
IMPACT This model's training methodology, which incorporates agent execution experience, could lead to more efficient and capable AI agents by better aligning model capabilities with real-world task requirements.
RANK_REASON First model release from a new startup founded by a prominent figure in the AI field (former head of Huawei's Noah's Ark Lab and Pangu model lead). [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Agentic Post-Training
- Huawei Noah's Ark Lab
- JiYuan LüDòng
- NeoHorse
- NeoHorse-1
- On-Policy Distillation
- OpenSquilla
- Peking University
- Qwen3.5 4B
- Routing-Guided Curriculum
- Routing Harness
- Tsinghua University
- Wuxin Qiong
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
