A free and secure coding AI, courtesy of China

DogFoodBuss writes: This week, Alibaba Qwen released Qwen3-Coder-Flash. It's believed to be the most capable AI coding model that can run locally on a developer laptop. The architecture delivers excellent throughput by only activating three of its 30-billion parameters per token generated, and supports 256k and 1M context sizes. Qwen also claims that their model can beat GPT 4 and DeepSeek V3 in many benchmarks.

Ready-to-use quantizations are now available from Unsloth AI https://huggingface.co/unsloth... and in Apple's MLX DWQ format on HuggingFace https://huggingface.co/mlx-com...