Ollama released v0.32.10-rc1, a new release candidate, introducing several enhancements. A notable change is the default repeat penalty for models that do not explicitly set this parameter; it now defaults to 1.0 (off) instead of 1.1. This adjustment aligns Ollama's behavior with other inference engines and is expected to speed up speculative decoding. The update also includes performance improvements, specifically a 7-8% faster prefill for certain NVFP4 MLX models, with gains seen on Qwen3.6 and Muse Glimmer models due to a global scale optimization. Additionally, a bug preventing blob verification from being skipped when an OCI manifest's config and layer shared a digest has been fixed.