Lab
Local-model tests and contributed benchmark runs, on hardware we actually have. Recipe write-ups live with the rest of the recipes; the raw files for a run stay here.
- Contributed run
~190 tok/swarm decode mean, thinking on and off
Weeyuga benchmark: Qwen3.8-27B on 2x R9700
A contributed run of the Weeyuga open benchmark: Qwen3.8-27B MXFP4 with a DFlash2 draft on two AMD Radeon AI PRO R9700s, cold and warm, three frozen prompts.
- Local-model test
15 vs 18 of 18runs passed, local Qwen 27B (thinking on) vs Sonnet 5.5 (adaptive)
A 27B model on two home AMD cards passed 15 of 18. Sonnet passed 18.
We ran Qwen3.8-27B on two Radeon AI PRO R9700s against Claude Sonnet 5.5 and Haiku 5.5 on six auto-graded tasks. Sonnet with thinking passed all 18 runs for $0.30. Qwen with thinking passed 15 for $0 in API, and most of its git patches didn't apply.