Ornith-1.0 open-weight agentic-coding models released

Category: AI
Status: discussion
Sources: GitHub repository, discussion
Summary: deepreinforce-ai released Ornith-1.0, a family of agentic-coding models (9B, 31B dense, 35B-MoE, and 397B-MoE) fine-tuned from Gemma 4 and Qwen 3.5 with a reinforcement-learning training loop that generates its own solution rollouts and task-specific harnesses; the 9B variant fits on a single 80 GB GPU. The "self-improving" label refers to the training method, not runtime self-modification.
Comments: Commenters noted the 31B dense variant lacks published weights or benchmarks, that on at least one cyber test it found only the bug nearly every model finds and degraded sharply without tool access, and reported it underperforming base Qwen3.6-27B in practice.
Why it matters: It is another open-weight agentic-coding entrant, but the distance between its framing and reproducible results illustrates the verification burden these self-reported model claims place on practitioners.