What's Fixed

Ollama v0.17.3 resolves a parsing issue affecting Qwen 3 and Qwen 3.5 model families where tool calls would not be recognized correctly when emitted during the model's thinking phase.

Impact

This fix is important for users leveraging Qwen's extended reasoning capabilities with tool use. Previously, models like Qwen 3 that emit tool calls while thinking would have those calls silently dropped or misprocessed. Now the runtime correctly captures and processes these tool invocations regardless of when they're emitted in the thinking process.

Who Should Update

Developers using Ollama with Qwen 3 or Qwen 3.5 models, particularly those relying on agentic workflows with function calling, should upgrade to benefit from this fix.

Download: Available now on GitHub releases