Ornith 1.0 9B
deepreinforce-ai/Ornith-1.0-9B
published Jun 2026 · updated Jun 2026
Ornith 1.0 9B is a text-generation model for agentic coding tasks, using a self-improving reinforcement learning framework to jointly optimize scaffolding and solution rollouts.
specs
| Task | Text Generation (Agentic Coding) |
| Architecture | Dense Transformer |
| Parameters | 9 Billion |
| License | MIT |
about this model
Ornith-1.0-9B is a text-generation model designed for agentic coding tasks, optimized for single-GPU deployment and achieving state-of-the-art results among open-source models of comparable size.
It is the most lightweight member of the Ornith family, which also includes 31B-Dense, 35B-MoE, and 397B-MoE variants post-trained on Gemma 4 and Qwen 3.5. The model employs a self-improving reinforcement-learning framework that jointly optimises both the scaffolding and the solution rollouts, enabling the discovery of better search trajectories and higher-quality code solutions. It is released under the MIT license.
Benchmarks
The table below compares Ornith-1.0-9B with Qwen3.5-9B, Qwen3.5-35B, Gemma4-12B, and Gemma4-31B on several agentic coding benchmarks.
| Ornith-1.0-9B | Qwen3.5-9B | Qwen3.5-35B | Gemma4-12B | Gemma4-31B | |
|---|---|---|---|---|---|
| Agentic Coding | |||||
| Terminal-Bench 2.1 (Terminus-2) | 43.1 | 21.3 | 41.4 | 21.0 | 42.1 |
| Terminal-Bench 2.1 (Claude Code) | 40.6 | 18.9 | 38.9 | - | - |
| SWE-Bench Verified | 69.4 | 53.2 | 70.0 | 44.2 | 52.0 |
| SWE-Bench Pro | 42.9 | 31.3 | 44.6 | 27.6 | 35.7 |
| SWE-Bench Multilingual | 52.0 | 39.7 | 60.3 | 32.5 | 51.7 |
| NL2Repo | 27.2 | 16.2 | 20.5 | 10.3 | 15.5 |
| Claw-eval Avg | 63.1 | 53.2 | 65.4 | 32.5 | 48.5 |
| SWE Atlas - QnA | 17.9 | 9.2 | 13.2 | - | - |
| SWE Atlas - RF | 16.6 | 4.3 | 10.2 | - | - |
| SWE Atlas - TW | 15.3 | 4.4 | 9.8 | - | - |
As a reasoning model, Ornith-1.0-9B outputs a <think> block before the final answer, supporting chain-of-thought during agentic tasks.
best for
- ·Automated bug fixing and code repair via SWE-bench tasks
- ·Terminal-based agentic coding workflows with tool use and scaffolding
- ·Efficient single-GPU deployment for repository-level code generation
FAQ
It is best for agentic coding tasks such as automated bug fixing, repository-level code generation, and terminal-based agentic workflows.
It outperforms Qwen3.5-9B and Gemma4-12B on coding benchmarks: SWE-Bench Verified (69.4% vs 53.2% and 44.2%) and Terminal-Bench 2.1 (43.1% vs 21.3% and 21.0%).
It is released under the MIT license with no regional restrictions, free for global use.
Use the gigarouter OpenAI-compatible endpoint with your API key. The model accepts standard chat messages and returns a reasoning block inside <think> tags.
It is a reasoning model: by default the assistant response begins with a <think> chain-of-thought block, followed by the final answer. The API supports chat completion requests with a reasoning parser.
We're benchmarking and onboarding Ornith 1.0 9B as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.