Agents & InferenceSimon Willison

Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

A new MIT-licensed open-weights model family (9B/31B dense, 35B/397B MoE) built on Gemma 4 and Qwen 3.5 hits SOTA among comparable open-source models on coding, and the 35B Q4 quant runs locally in ~20GB at 103 tok/s while handling multi-step agentic tool calls reliably. This means you can run a competent self-scaffolding coding agent on a single workstation GPU under a permissive license—no API costs, no per-token billing, and no rate limits for your harness.

LinkedIn

Two AI summaries of each story, blind-voted — see today's agents & inference digest →