Liquid AI finetune beats GPT6 Astra at chess, 100x faster per move, fully local
avapamini · x · 2026-11-11
Developer manoletre finetuned Liquid AI's d1-omni-600b into a chess model that listens to voice commands, then pitted it against GPT6 Astra with reasoning enabled — Astra took over 100x longer per move on average.
A notable part of the process was using Claude to run optimization experiments: Claude found only 2k tokens were actually used, so a large portion of the embeddings matrix could be removed, shrinking model size, peak memory, and move latency while keeping chess strength constant. The final model runs entirely locally in a browser tab on the author's website.
More from Fun
- Meme: While plague raged in Europe, Newton was just like 'haha colours' — miniapeur · 2026-11-11
- Alien asks to 'be real for a second', gets a mirrored-z joke — miniapeur · 2026-11-11
- Artist with disability releases his artwork as a dataset for anyone to remix — plzno1 · 2026-11-11
- Solo dev builds helicopter-vs-medieval-castle browser game entirely with Fable 5.2 — builderjaydub · 2026-11-11
- AI virtual act fAIkout adds new cast member Roy to Rawkhaus Records — AIandDesign · 2026-11-11
- GTA V on Specs glasses: light trails and rain leap off the big screen into your room — stspanho · 2026-11-11