Local Ornith 1.5 35B test: Model solid, hardware the bottleneck
DanWahlin · x · 2026-08-27
Testing Ornith 1.5 35B (Q5) locally in GitHub Copilot on an M3 Max showed solid model performance and tool use. However, quantized local inference remains bottlenecked by hardware for speed and quality in agentic tasks.
More from Apps
- Splat.js: Browser-based 3D Gaussian Splatting trainer — petewoodbridge · 2026-08-27
- Measuring AI Performance with Custom Evals like KateBench — every · 2026-08-27
- SurveyorBench evaluates LLMs on civil engineering workflows — ycombinator · 2026-08-27
- Colony Game Launches AI-Powered 3D Item Generator — templecrash · 2026-08-27
- Google's Product Architecture Criticized for Confusing Users — HaktanSuren · 2026-08-27
- Grok Bot resets free usage limits for all users — Baconbrix · 2026-08-27