← OpenAI
Creator Magic
Local AI That Codes Like Claude - Ornith 1.0
⏱ 32 min video · 2 min read30 Jun 2026
TL;DR
Creator Magic host tests Ornith 1.0 35B, a local AI model claimed to rival Claude Opus, by using it inside Open Code on a Mac Studio M3 Ultra to build a 3D flight simulator and a Kanban board. Results are janky but partially functional, revealing both the promise and current limitations of local AI for coding tasks.
Key points
1
Ornith 1.0 35B is a local AI model from a US lab, claimed to rival Claude Opus in coding benchmarks, and can run on Apple Silicon with sufficient unified memory (32GB+ recommended for the 35B version).
2
The model was tested via Open Code as a harness, running inference locally on a Mac Studio M3 Ultra, generating a 3D flight simulator in Three.js and a Kanban board — both partially worked but were buggy.
3
The flight simulator produced a corrupted HTML file with two documents merged together; Claude Code was needed to diagnose and partially fix it, ultimately producing a janky but playable result with 6 biomes.
4
The host argues the ideal workflow is a hybrid: frontier models like Claude Opus set direction and verify output, while local models like Ornith or Qwen 3 do the heavy coding, reducing cloud dependency.
5
Ornith models are available from Deep Reinforce AI in sizes from 9B (runs on most MacBooks) to 397B (requires 256GB unified memory or high-end Nvidia GPU like RTX 3090/4090).
Key takeaways
→
Run Ornith 1.0 9B on a standard MacBook or Mac Mini for a low-barrier entry into local AI coding; step up to 35B if you have 32GB+ Apple Silicon RAM.
→
Use local models for iterative coding tasks but pair them with a frontier model (e.g. Claude Opus) as a verifier/orchestrator to catch corrupted or incomplete output.
→
Download Ornith models from Deep Reinforce AI; the 35B quantized version is a practical sweet spot between capability and hardware requirements for most enthusiasts.
Notable quotes
“You can take the boy out of radio, but you can not take the radio out of the boy.”
“Local AI needs direction. Imagine Ornith does a bit of janky work, Claude Opus verifies that janky work, and goes back to Ornith and says yeah you messed up dude, can you fix it.”
“The big rally cry is go for local AI because we never know when models can be pulled from underneath us.”
Worth watching?
⏭️
Worth watching the full video?
Skip the video — it is a raw, often-interrupted live stream with significant time spent on streaming tech issues; the key findings on Ornith 1.0 performance and hardware requirements are all captured here.
Saved you some time? The creator still deserves a like.
Want this for your own channels?
Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.
Try it free