Qwen: Qwen3 235B A22B

Qwen3-235B-A22B is a state-of-the-art 235B parameter Mixture-of-Experts (MoE) model developed by Qwen. It intelligently activates 22B parameters per forward pass, offering a unique dual-mode operation: a 'thinking' mode for intricate tasks like complex reasoning, advanced mathematics, and code generation, and a 'non-thinking' mode optimized for general conversational efficiency. This flexibility makes it highly adaptable to diverse application needs. This model demonstrates exceptional reasoning capabilities, robust multilingual support across over 100 languages and dialects, advanced instruction-following, and sophisticated agent tool-calling functionalities. It natively handles a substantial 32K token context window, which can be further extended up to an impressive 131K tokens using YaRN-based scaling, enabling it to process and understand extensive information. With a maximum output of 4K tokens, it's ideal for detailed responses. Qwen3-235B-A22B supports functions, code, and streaming capabilities. Pricing is competitive at $0.20 per 1M input tokens and $0.60 per 1M output tokens. It's best suited for chat, code generation, and mathematical problem-solving. Please note, it does not support image generation or internet access.