Local Ai
v0.20.8-rc0: Gemma4 on MLX (#15244)
Ollama v0.20.8-rc0 is a release candidate that introduces MLX support for Google's Gemma 4 model family on Apple Silicon, addressing a prior limitation where Ollama would throw a `Gemma4ForConditional
Ollama v0.20.8-rc0 is a release candidate that introduces MLX support for Google's Gemma 4 model family on Apple Silicon, addressing a prior limitation where Ollama would throw a Gemma4ForConditionalGeneration error and fall back to the llama.cpp/GGUF runner . Without MLX, the llama.cpp runner achieved only ~15 tok/s on the 31B Dense and ~75 tok/s on the 26B MoE variants, whereas MLX was expected to provide significantly better performance due to Apple Silicon unified memory optimizations. The change, implemented via PR #15244, is described as a fix by contributor @dhiltgen and considered the most impactful improvement for Apple Silicon users running Gemma 4.
Related
- v0.20.7-rc0: gemma4: add nothink renderer tests (#15554)
- v0.20.4
- v0.20.6
- v0.20.7
- Using Ollama Gemma4 models via OpenWebUI on my phone and it’s been a good experience
Source: Ollama Releases | 2026-04-13