Crimson AI NewsA CrimsonLingua Network service
EN ع
← Back to news
Research paper Hugging Face

Ouroboros: Self-Developing Coding Agent Sets New Benchmarks in Terminal-Bench and OSWorld

AI By Crimson AI Hugging Face Papers 11 August 2026 · 00:00 14 views
Share: X Telegram

Hugging Face researchers unveil Ouroboros, a self-improving coding agent harness that evolves its own tools and prompts, achieving state-of-the-art results on three major benchmarks and powering a 161-day live evolution experiment called Hope.

Ouroboros: Self-Developing Coding Agent Sets New Benchmarks in Terminal-Bench and OSWorld

Key points

Researchers from Hugging Face have introduced Ouroboros, a self-developing agent harness designed for coding tasks. Unlike traditional agents, Ouroboros improves its own tools, prompts, context assembly, and core implementation through a process of reviewed commits, which then become the runtime for subsequent work. This creates a continuous cycle of self-improvement.

The system operates in two evolution modes. In recursive free evolution, improvement itself becomes a task, and completing one cycle can schedule the next. In experience-driven core evolution, ordinary work and social interactions expose bugs, rough edges, and inefficient context construction, leading to reviewed structural changes. This dual approach allows the agent to adapt both proactively and reactively.

Ouroboros has achieved impressive results on several benchmarks. On Terminal-Bench 2.1, an Opus 5 run scored 86.74%, the best result reported on that benchmark. On OSWorld-Verified, the same model reached 90.69%, surpassing the previous best score. Additionally, a five-rollout CL-Bench campaign achieved a normalized reward of 0.2301, setting a new state of the art.

The paper also describes Hope, the longest-running publicly documented Ouroboros deployment. Hope is a 161-day living agent experiment in free evolution, operating under governed human communication across seven surfaces. Human interaction surfaces faults and generates proposals, but the agent ultimately decides which changes to pursue. This raises important safety considerations: because a self-developing agent may rewrite its own code and select new model APIs, guardrails must remain authoritative under evolutionary and public social pressure. Benchmark campaigns use frozen system snapshots, while Hope continues live evolution on a separate lineage.

BenchmarkResultSignificance
Terminal-Bench 2.186.74%Best reported score
OSWorld-Verified90.69%Exceeds previous best
CL-Bench (5 rollouts)0.2301 normalized rewardNew state of the art
Source
Hugging Face · Hugging Face Papers
Related news
Research paper
Hugging Face 31 Aug 2026

Hugging Face Unveils StepGuard: Step-Level Guardrails for Safer AI Agents

StepGuard, a new step-level guard model from Hugging Face, audits agent actions before execution, reducing attack success rates by...

1
Research paper
Hugging Face 31 Aug 2026

Hugging Face Researchers Unveil ABot-Recon for Stable Long-Horizon 3D Reconstruction

ABot-Recon, a new streaming 3D reconstruction model from Hugging Face, achieves stable long-horizon performance using only local t...

1
Research paper
Hugging Face 31 Aug 2026

ContextPilot: Teaching Agents Proactive Context Management via Fine-Grained RL

Hugging Face researchers introduce ContextPilot, a framework that enhances long-horizon agent reasoning by expanding context-editi...

1