CoEvo-RSI: When Harnesses Evolve, Which Models Should Follow?
In brief
- CoEvo-RSI synchronizes harness and model evolution through a meta-agent.
- Qwen3.5-9B, trained via supervised fine-tuning, rivals larger models.
- Outperforms DeepSeek-V4-Pro and GLM-5.3 on specific benchmarks.
Researchers introduce CoEvo-RSI, a recursive self-improving (RSI) framework that synchronizes the evolution of execution environments (harnesses) and models (brains). According to the announcement, the system uses a meta-agent to select or merge model versions based on historical execution archives, reducing mismatches between model capabilities and environment requirements.
The approach is tested on vertical tasks such as multi-agent failure attribution or smart-home instruction following. Results show that Qwen3.5-9B, trained solely via supervised fine-tuning on its own trajectories, competes with larger models, outperforming DeepSeek-V4-Pro and GLM-5.3 on specific benchmarks.
This summary gives you the essentials. The full article, its figures and quotes are on Hugging Face Daily Papers.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error