Ollama 0.40.7 is out
A new version of a popular tool for running models on your own computer.
Ollama 0.40.7, a new release of the tool for running language models on your own computer, adds pplx-decider, a 27-billion-parameter decision model from Perplexity fine-tuned from Qwen3.8-27B. The release also updates the bundled llama.cpp and MLX.
Most changes are fixes for people who connect other tools to Ollama. Conversations through its OpenAI-compatible Responses interface no longer fail when you switch between models with and without thinking, tool calls from clients such as Codex arrive in the right order, and requests routed to remote models on ollama.com stop when the client disconnects.
The ollama show command now reports which renderer and parser a model uses and where its chat template comes from. ollama launch claude no longer triggers a "window too large" warning, and more draft models work with speculative decoding.