How to Use Ornith 1.0: The Self-Improving Agentic Coding Model

What makes Ornith 1.0 different

Ornith 1.0 uses reinforcement learning to generate not just solution rollouts, but the scaffolding that drives those rollouts. By jointly optimizing the scaffold and the resulting solution, the model discovers better search trajectories and generates higher-quality code.

It achieves state-of-the-art results among open-source models of comparable size on Terminal-Bench 2.1, SWE-Bench, and NL2Repo benchmarks.

Install via Ollama

Ornith is available on Ollama. Run `ollama run ornith:9b` for the 9B variant or `ollama run ornith:35b` for the 35B variant.

The 397B variant requires multi-GPU infrastructure and is available via the DeepReinforce API.

Use for agentic coding tasks

Ornith excels at coding agent workflows where the model must plan, write, test, and iterate on code.

It performs particularly well on SWE-bench style tasks where the agent must understand a repository, identify the issue, and implement a fix.

The self-improving architecture means it gets better at scaffolding its own code generation process over time.

License and availability

MIT licensed — globally accessible with no regional limitations.

Available on Ollama, Hugging Face, and via the DeepReinforce API.

256K token context window on all variants.

FAQ

Use Ornith 1.0

1. Install Ollama: ollama run ornith:9b
2. Test with a coding task: generate a function with test cases
3. Try agentic mode: give it a repository-level task
4. Evaluate output quality for your use case
5. Scale to 35B if more capability is needed
6. Check HuggingFace for the latest fine-tuned variants