Ollama released version 0.40.8‑rc0 on 2026‑10‑21. The update ships a new model called Flint, aimed at coding agents that need to read unfamiliar code, trace bugs, add features, and write tests. Flint works with a 262,000‑token context window, so it can keep many files in memory at once. The first release candidate runs on MacOS; GPU support for other platforms will come later.
What Flint brings
Flint is an Apache‑2.0 licensed mixture‑of‑experts model. It has 40 billion total parameters but only about 2.5 billion are active for each token, which keeps inference fast. In the developer’s own tests, Flint scored 54.9% on Terminal‑Bench 2.1 and 74.2% on SWE‑bench Verified, two benchmarks that measure how well a model can write and fix code.
Why it matters
Developers using Ollama can now ask an AI to work across many source files without losing context, which may speed up debugging and feature development. The model is still in a release candidate, so performance may improve and broader hardware support is pending.
