Ornith-1.0-9B

About

Ornith-1.0-9B (ornith-ai/Ornith-1.0-9B) is the most lightweight member of the Ornith-1.0 family - a self-improving, dense 9B-parameter (post-trained on top of the Qwen 3.5 9B base) open-source model for agentic coding, designed for efficient single-GPU deployment. It is MIT licensed, globally accessible, and free from regional limitations.

Ornith-1.0's scaffold-rollout co-optimization training framework uses RL to learn to generate not only solution rollouts but also the scaffolds that drive those rollouts: by jointly optimizing the scaffold and the resulting solution, the model discovers better search trajectories and generates higher-quality solutions. The 9B model achieves state-of-the-art results among open-source models of comparable size on agentic coding benchmarks (Terminal-Bench 2.1, SWE-bench Verified/Pro/Multilingual, NL2Repo, Claw-eval, SWE Atlas), outperforming its Qwen3.5-9B base (e.g. 69.4 vs 53.2 on SWE-bench Verified). It natively emits imdi ... dimdi reasoning blocks and supports tool calling; GGUF builds are published at deepreinforce-ai for llama.cpp, Ollama, OpenClaw, OpenHands, Hermes and opencode integration. Context length 262,144 tokens (native; no extension documented).

Training Data Post-trained (RL, self-scaffolding) on top of Qwen 3.5 9B base. RL jointly optimizes scaffold and solution rollouts for agentic coding.