Qwen3 235B A22B

Qwen3-235B-A22B is a flagship mixture-of-experts model from Alibaba with 235 billion total parameters and 22 billion activated per token, drawing on 128 experts with 8 active at a time. This sparse design keeps inference cost low while preserving broad capability. The model supports switching between a thinking mode for complex mathematical, logical, and coding tasks, where it generates step-by-step reasoning in think blocks, and a non-thinking mode for fast conversational responses. It performs strongly on mathematics, coding, and general knowledge, and handles 100+ languages, drawing on roughly 36 trillion tokens of pretraining data. Native context covers around 40K tokens, extensible toward 131K with YaRN, balancing the depth of reasoning models with the speed of dense models. This FP8 build uses fine-grained quantization with a 128 block size to shrink the memory footprint. Open-weight under Apache 2.0.