@unslothai
Run and train models locally with the Unsloth Desktop app. 🦥 https://github.com/unslothai/unsloth
Technology
Alibaba's Qwen team reportedly released Qwen4-Max, an MoE model with native vision and a 1 million-token context window, priced below the prior Max tier. Smaller open-weight Qwen4 models were teased, and early Arena placements were reported as strong.
AI
In NVIDIA's Reachy Mini demo, Bolt handles conversation while Cosmos 3 Edge supplies vision on Jetson Orin Nano 8GB.
AI
Its developer says Flint exceeds 125 tokens per second on an M4 Pro MacBook Pro at short contexts.
AI
finkd says the model runs well on new Mac Studios, crediting work with Apple and John Ternus.
AI
UnslothAI says the models work on CPU, Mac, Windows, Linux and GPU setups. It also says Unsloth Desktop can serve Laya through a Jev-compatible API.
AI
Unsloth says its Docker image works on NVIDIA and AMD and offers a new graphical interface or a notebook workflow, with no setup required.
AI
Retweet by Unsloth AI co-founder shares local speed claims for Qwen model.
AI
The 27B model is now available on Ollama for agentic coding and professional tasks.
AI
Users can run and train models on Mac, Windows, and Linux hardware.
AI
Open 30B MoE model with 3B active parameters targets always-on agents.
AI
Mark Zuckerberg posted that Meta opened weights for a local model with another release planned soon.