Ollama adds Claude Code auto-install

Ollama, the tool for running open models on your own machine, shipped version 0.30.11 on June 25. The change builders will notice: it now auto-installs Claude Code and opencode when they are missing, so you can drive a coding agent against a local model without setting up the agent separately. The release also adds thinking-capability detection for opencode, which lets it tell whether the local model it is talking to supports a reasoning mode.

The rest is plumbing, but useful plumbing if you are on Windows. Ollama now uses the host Vulkan loader on Windows and fixes a bug that had inverted integrated and discrete GPU classification on hybrid-graphics laptops, a common cause of the wrong chip getting the work. It also adds the sm_86 CUDA architecture to its Windows build, widening the set of NVIDIA cards it accelerates. This stable release promotes the features that were sitting in last week’s release candidate.

What it means for you

Nothing here changes what a model can do; it lowers the friction of running one locally and wiring a coding agent to it. That matters if you are weighing local open models against a hosted API for cost or privacy reasons. The open side keeps moving fast, from GLM-5.2 topping the open-weights leaderboard to Kimi K2.7-Code shipping leaner reasoning. If you are setting this up for the first time, our guide on coding agents for non-developers walks through the choices.