# Ollama v0.40.5 release candidate adds CUDA kernels

> Ollama’s v0.40.5-rc0 includes CUDA ports of Metal custom kernels and changes to Responses API tool-call history. It is a release candidate.

Oossa · 2026-10-14 · https://oossa.com/en/ollama-v0-40-5-release-candidate-adds-cuda-kernels

Ollama published v0.40.5-rc0, a release candidate for its local AI software. The update ports custom kernels previously used with Apple’s Metal graphics technology to CUDA, NVIDIA’s platform for GPU computing.

It also lets the OpenAI-compatible Responses API accept custom tool calls in conversation history. The release notes do not give performance figures or say when a final v0.40.5 release will arrive.

## The facts

- Version: v0.40.5-rc0, with a CUDA port of Metal custom kernels.
- Published: October 14, 2026.

## Why it matters

For Ollama users running NVIDIA GPUs, the CUDA port may make these custom kernels available on their hardware, though the notes do not report performance or confirm what will change in the final release.

## Sources & references

1. [ollama/ollama v0.40.5](https://github.com/ollama/ollama/releases/tag/v0.40.5-rc0) – Ollama, 2026-10-14

Last updated: 2026-10-14