A new GPT model just dropped for Apple Silicon. The gpt-oss-20b-tq3 is a 20B parameter MoE text generator, optimized with TurboQuant and MLX for fast, local inference on your Mac.

Build your own conversational AI or text generation app right on your laptop. This model is perfect for chatbots, creative writing, and coding assistants, all running locally without cloud costs.

It uses a mixture of experts (MoE) architecture with 20B total parameters. The TurboQuant technique shrinks model size while maintaining quality, and safetensors ensure safe, fast loading.