Qwen3-235B-A22B-Instruct-2507 is a multilingual, instruction-tuned mixture-of-experts language model based on the Qwen3-235B architecture, with 22B active parameters per forward pass. Optimized for general-purpose text generation, including instruction following, logical reasoning, math, code, and tool usage. Supports a native 262K context length and delivers significant gains in knowledge coverage, long-context reasoning, and coding benchmarks.
Not enough TTFT data