●Change8

v0.11.0

📦 ollama
✨ 8 features🐛 1 fixes🔧 5 symbols

Summary

Ollama v0.11 introduces support for OpenAI's gpt-oss models (20B and 120B) featuring native MXFP4 quantization, agentic capabilities, and configurable reasoning effort.

Migration Steps

  1. Download and install Ollama version v0.11 from the official website.
  2. Run 'ollama run gpt-oss:20b' or 'ollama run gpt-oss:120b' to pull the new models.

✨ New Features

  • Support for OpenAI gpt-oss 20B and 120B open weight models.
  • Native support for MXFP4 (4.25 bits per parameter) quantization format.
  • Built-in web search tool for model augmentation.
  • Support for native function calling, python tool calls, and structured outputs.
  • Full chain-of-thought access for model reasoning transparency.
  • Configurable reasoning effort levels (low, medium, high).
  • New optimized kernels for the MXFP4 format in the Ollama engine.
  • kvcache: Enable Sliding Window Attention (SWA) to retain additional entries.

🐛 Bug Fixes

  • kvcache: Added logging for cache contents when a slot cannot be found.

🔧 Affected Symbols

gpt-oss:20bgpt-oss:120bMXFP4kvcacheSWA