qwen

drops a family of small models. quite interesting to try.

the small model family ranges from 0.8B to 35B parameters. The 0.8B model is an interesting one; an efficient way to play on the local computer. It is surprisingly efficient for daily duties; organizing files, classifying things, etc.

All the qwen small models family supports vision.

And the largest model of this family is a mixture of experts model, 35B in total but running 3B on a single task.

In my opinion, this family of small models clearly target the local deployment field as agentic tools in the local surge to climb with coding tools and claw variants.

May want to give it a shot. Runs in nearly all devices.