v1.6.0
LatestA new local model, and a model picker that only offers what your Mac can actually run.
Highlights
Qwen3.8 27B, on your Mac
Run Qwen3.8 27B locally in two builds: the standard one for a 32 GB Mac, and a smaller-quantization build that brings the same model to a 24 GB Mac, trading some quality for the fit. Both accept images and think briefly before answering.
A model picker that knows your Mac
Creating an assistant now shows only the local models your machine has the memory to run once the assistant's VM has taken its share, along with the smallest Mac each model needs. Turn the filter off to see the whole catalog, with quantization labels and a clear marker on anything that will not fit.
- Feature Run Qwen3.8 27B locally, in a 32 GB build and a 24 GB smaller-quantization build. Both accept images.
- Feature See only the local models your Mac can run when creating an assistant, with a toggle to show the whole catalog.
- Feature Run local models that accept text only, with image attachments disabled in the composer instead of failing mid-reply.
- Fix Fixed new assistants still running the assistant's bundled browser in the background after switching to the browser-navigation skill.
- Other Updated the bundled local-inference engine.