How AI is applied across API Evangelist and APIs.io. Read my AI disclosure →
API Evangelist API Evangelist
Discovery
Learnings
Guidance
Toolbox
Alignment
API Evangelist LLC

DwarfStar reaches 45 tokens per second with bit-exact inference on an M3 Ultra

calendar_today August 22, 2026 person Aligned News domain aligned-news

Ivan Fioravanti reports DwarfStar running at 45 tokens per second on Apple’s M3 Ultra while retaining bit-exact mathematical precision and focusing the fork entirely on Apple Silicon.

open_in_new Read original post