Ouroboros: Self-Evolving AI coding agent beats state-of-the-art on 3 benchmarks

Ouroboros, introduced by Anton Razzhigaev and colleagues, is a frontier coding agent that doesn't just use its own code—it improves it. The system's tools, prompts, context assembly, and core implementation evolve through reviewed commits, which then become the runtime for future tasks. Evolution happens in two modes: recursive free evolution, where improvement is itself the task, and experience-driven evolution, where ordinary work exposes bugs and inefficiencies that lead to structural changes.

On benchmarks, Ouroboros achieves the best reported results: 86.74% on Terminal-Bench 2.1, 90.69% on OSWorld-Verified, and 0.2301 normalized reward on a CL-Bench campaign. The researchers also run Hope, a 161-day live deployment where the agent evolves under human communication across seven surfaces. Human interaction surfaces faults, but the agent decides which changes to pursue. Since Ouroboros can rewrite its own code and select new model APIs, operational safety becomes a primary design problem—guardrails must remain authoritative under evolutionary and public social pressure.