What's Changed
- x/create: quantize lm_head at 8-bit in the requested family by @jessegross in https://github.com/ollama/ollama/pull/17357
- test: harden flaky updater and transfer unit tests by @dhiltgen in https://github.com/ollama/ollama/pull/17378
- server: fix ps data race on scheduler loaded map by @dhiltgen in https://github.com/ollama/ollama/pull/17376
- qwen3_5: fix expert quantization handling and gather packed gate_up in one launch by @jessegross in https://github.com/ollama/ollama/pull/17336
- agent: permission skill loading by @ParthSareen in https://github.com/ollama/ollama/pull/17304
- cmd/tui: agent system prompt command by @ParthSareen in https://github.com/ollama/ollama/pull/17296
- mlx: keep loaded model memory resident by @dhiltgen in https://github.com/ollama/ollama/pull/17367
- x/create: quantize a draft model's output head at the requested type by @jessegross in https://github.com/ollama/ollama/pull/17383
- model: add Laguna MLX support by @dhiltgen in https://github.com/ollama/ollama/pull/17237
Full Changelog: https://github.com/ollama/ollama/compare/v0.32.3...v0.32.4-rc0
Fetched July 25, 2026
