FULL STORY
Gemma 4 MLX Challenge Pushes Mac Inference
A community MLX challenge to optimize Gemma 4 inference on Apple Silicon pushed speeds from 178 tok/s to nearly double within 48 hours.
2026-08-28 ~ 2026-08-30 · 2 episodes · 4 posts
Episode 1 · Gemma 4 MLX Challenge Boosts Mac Inference to 178 tok/s (2026-08-28, 2 posts)
Darkbloom launched the Gemma 4 MLX Challenge to optimize Gemma 4 inference on Mac. The best entry has lifted throughput from a 153.1 tok/s baseline to 178 tok/s, an 8% gain driven mainly by faster decoding.
- Gemma 4 MLX Challenge launches with 8% speedup on Mac — gajesh · 2026-08-28
- Gemma 4 MLX Challenge Boosts Local Inference to 178 tok/s — TheMoonMidas · 2026-08-28
Episode 2 · Gemma 4 MLX Challenge: Near 100% Performance Gains in 48 Hours (2026-08-29, 2 posts)
In the Gemma 4 MLX Challenge, the 26B model's performance on Apple Silicon improved nearly 100% within 48 hours, while Gemini 3.7 Flash topped the leaderboard at 71.1%, beating Claude Fable 5.
- Gemini 3.7 Flash Beats Claude in Gemma 4 Mac Optimization Challenge — minsuk_chang · 2026-08-29
- Gemma 4 26B MLX Challenge Nears 100% Performance Gain in Under 48 Hours — gajesh · 2026-08-30