Simon Willison
8/10 signal
Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding
agentictool-use
What happened
DeepReinforce released Ornith-1.0, an open-weights (MIT licensed) model family (9B, 31B, 35B MoE, 397B MoE) built on Gemma 4 and Qwen 3.5. It is specifically optimized for agentic coding and self-scaffolding. Early testing by Simon Willison confirms the 35B MoE model performs exceptionally well at running multi-step tool-use loops and codebase search tasks locally.
Why it matters
It provides a highly capable, permissively licensed open-weights alternative for driving complex agentic developer loops locally.
The take
This is a highly actionable release. Models explicitly fine-tuned to handle the tool-calling and self-correction loops of agentic scaffolds (rather than just raw code generation) are exactly what the community needs to build reliable local coding agents.
Do this
Download the GGUF version of Ornith-1.0 (such as the 35B MoE) and test it in your local agentic coding pipelines or tool-calling harnesses.
Don't read this site daily. Get it in your inbox.
The daily brief and Sunday deep dive — distilled, scored, and opinionated. For builders only.