v0.32.6
Breaking Changes📦 ollamaView on GitHub →
⚠ 1 breaking✨ 3 features🐛 4 fixes🔧 4 symbols
Summary
This release enhances performance for Qwen3.5 on Apple GPUs and standardizes the `/v1/chat/completions` streaming API to match OpenAI's format. It also includes several TUI improvements and bug fixes.
⚠️ Breaking Changes
- Experimental image generation has been temporarily removed. Users requiring image generation should continue using version 0.32.5.
✨ New Features
- Qwen3.5 is faster on Apple GPUs due to the MLX engine automatically using the model's MTP head for speculative decoding.
- The `/v1/chat/completions` streaming API now aligns with OpenAI's wire format, with `role` only on the first chunk, `finish_reason` on its own chunk, and usage in a separate chunk when `stream_options.include_usage` is enabled.
- The `ollama run kimi-k3` command now supports `kimi-k3:cloud` for cloud-only models that do not publish a default tag, preventing previous failures.
🐛 Bug Fixes
- Truncated OpenAI responses now correctly report `finish_reason: "length"` instead of `"tool_calls"`.
- TUI fix: Pipe-delimited prose will no longer render as a table.
- TUI fix: Pressing Enter now accepts the highlighted `@` file completion.
- TUI fix: Scrolling in `/prompt` is no longer laggy.