Ouroboros: Self-Developing Coding Agent Tops Multiple Benchmarks
Anton Razzhigaev · hf · 2026-08-11
Ouroboros is a self-developing frontier coding agent capable of autonomously improving its tools, prompts, context assembly, and core implementation through reviewed commits, which then become the runtime for subsequent tasks.
The agent features two core evolution modes:
- Recursive Free Evolution: Treats self-improvement as a task, where completing one evolution cycle can automatically schedule the next.
- Experience-Driven Core Evolution: Leverages bugs and inefficient context construction exposed during ordinary work and social interactions to trigger structural code changes.
Performance-wise, Ouroboros sets new state-of-the-art records across multiple benchmarks: scoring 86.74% on Terminal-Bench 2.1, 90.69% on OSWorld-Verified, and achieving a normalized reward of 0.2301 on CL-Bench. The project also details Hope, a 161-day living agent experiment in free evolution that autonomously decides which code changes to pursue based on human interaction, while maintaining strict operational guardrails.
More from coding & agent
- CMU Launches ExploitBench: Testing AI Agents on Real V8 Exploitation — cyb3rops · 2026-08-11
- Zhipu's ZCode Hits 1M Users: GLM Coding Plan Resets Limits, Boosts Multi-Agent Collaboration — jietang · 2026-08-11
- 4x DGX Spark Cluster Achieves 44.6 tok/s on GLM-5.2 for Real Agent Workloads — EAccelerate_42 · 2026-08-11
- Developer Calls for AI Agent Platforms to Expose Cost Control APIs — arthurcolle · 2026-08-11
- Why Are AI Agents Still Fragile in Real-World Workflows? — No_Progress92 · 2026-08-11
- Agent Memory Distillation boosts small LLM agents by up to 27.2% accuracy via hierarchical teacher memory — kaist-ai · 2026-08-11