Ollama released v0.40.4 on October 12, 2026, one day after v0.40.3. The headline change is decision models in the CLI; the release lists no breaking changes. Read the release notes.
Decision models in the terminal
Decision models can now be run directly with ollama run. The release example asks a model to label a support ticket:
ollama run nimble --ask "Is a refund being requested?" < ticket.txtUse cases include yes or no questions, classifying text into labels, and rubric-based scoring. Kev decision models are also supported by the MLX runner on Apple silicon, and llama.cpp was updated with native decision model support.
Other changes
ollama launchcan now use usage credits without upgrading to a paid plan.- The Codex launcher model picker now lists locally installed Ollama models, switchable with
/model. - Inference is faster when verified manifests and already-loaded models can be reused.
- Multimodal embeddings are supported through llama-server.
Fixes and performance
The release fixes loading multimodal embedding models with small context windows, improves Gemma 4 prefill performance on NVIDIA GPUs under the MLX runner, clips long Laya states instead of rejecting them, and improves speculative decoding reliability, including recurrent state handling after rollback. Documentation wording around privacy and cloud services was also clarified.
This report covers the October 12 release notes. Our source verification guide describes how we read them.