Sam Witteveen · 2026-06-26 · notable
Sam Witteveen: 'Introducing Ornith 1.0' — open-weight coding LLM walkthrough
Sam Witteveen walks through Ornith 1.0, DeepReinforce's MIT-licensed coding model family whose RL loop learns its own scaffold — uploaded hours after the 9B–397B weights landed on Hugging Face.
A first hands-on look at Ornith 1.0 — DeepReinforce's open-weight coding LLM family that trains its own RL scaffold.
What is it?
Sam Witteveen breaks down Ornith 1.0, the MIT-licensed agentic coding model family DeepReinforce released on June 25, 2026. The lineup spans a 9B dense, a 31B dense, a 35B mixture-of-experts, and a 397B MoE flagship, all built on Gemma 4 and Qwen 3.5 pretrained checkpoints.
How does it work?
The walkthrough explains the self-scaffolding RL loop at the core of Ornith 1.0: the model first proposes a task-specific scaffold, then uses that scaffold to generate a solution rollout, with both targets co-evolving during training. Sam Witteveen is one of the eight AI YouTubers AI/TLDR tracks for fast, hands-on coverage of open frontier releases.
Why does it matter?
Coding-model news tends to land as a paper plus weights and little else. Ornith 1.0 is open-source and hits 82.4% on SWE-Bench Verified, but most engineers will first meet it through Sam Witteveen's video — a same-day demo of the variants, the RL recipe, and what the scaffold-co-evolution buys over a fixed harness.
Try it
https://www.youtube.com/watch?v=uD4-uy0GmHE