Ollama Releases v0.40.7 with New 27B Model and Bug Fixes

What happened

Ollama released version v0.40.7, featuring a new 27B parameter decision model named pplx-decider, which is fine-tuned from Qwen3.8-27B. The update also fixes bugs related to OpenAI-compatible Responses and streaming requests, ensuring better reliability.

Additional improvements include the ability to switch between models during a conversation without causing failures, and the cancellation of remote model requests if the client disconnects, preventing unnecessary computation.

Why it matters

These updates enhance the stability and performance of Ollama's model interactions, particularly for developers using OpenAI-compatible APIs. The new model and bug fixes contribute to a more robust and efficient user experience.

Key details

The pplx-decider model is a 27B parameter decision model fine-tuned from Qwen3.8-27B. Ollama also updated its llama.cpp and MLX components to newer versions, improving overall system compatibility.

The release includes a full changelog detailing the v0.40.6 to v0.40.7 update, available on GitHub. The 'ollama show' command now provides more detailed information about model metadata, including renderer, parser, and chat template sources.