Ollama has published version 0.40.12-rc0, a pre-release of its software for running AI models. The update adds DFlash2 support and includes an MLX version bump that Ollama says speeds up prompts for Gemma, Flint and Laguna.
It also changes how MLX handles small images for compact Gemma 4 vision towers, aiming to preserve detail, and updates its llama.cpp version to b11901.
Why it matters
People testing Ollama’s pre-release can try DFlash2 support and the listed MLX changes, but the release notes do not quantify speed gains or say when a stable release will arrive.
