Ollama v0.40.12-rc0 pre-release adds DFlash2 in MLX and Gemma 4 vision tweaks

caolonghua · x · 2026-10-27

Ollama shipped the v0.40.12-rc0 pre-release on October 27, with three notable changes:

• DFlash2 support added to MLX
• Small-image detail preservation for compact Gemma 4 vision towers, plus an MLX/MLX-C version bump that speeds up prompting for Gemma-family models
• llama.cpp updated to b11901

The release notes list the changes without quantifying their practical impact, and as a pre-release it should be treated as untested for stability.

Original post →