What changed
Ollama’s 0.40.4-rc1 release candidate, published October 12, includes several builder-facing changes. The notes update llama.cpp to b11521, add EmbeddingGemma2 support, improve server reuse of verified manifests and loaded models in inference routes, and fix issues affecting MLX.
The most visible workflow change is that installed local models now appear in Codex’s `/model` picker. That lowers friction for developers who want to move between a local model and hosted options without manually reconstructing model names and endpoints.
What the result actually shows
These are infrastructure details, but they change the everyday loop: model selection, loading and serving are all places where local experiments can become annoying. Reusing a verified manifest and already-loaded model can avoid repeated setup work; support for a new embedding model gives retrieval experiments another local option.
Important label: this is an `rc1` prerelease, not the stable release. Treat it as a test build and check the release notes for platform-specific behavior before upgrading a working setup.
Source published 2026-10-12. Coverage is based on the maker’s announcement and demonstration.