Alibaba Unveils Qwen4-Max with Native Vision and 1M Token Context
Alibaba unveiled its flagship language model, Qwen4-Max, featuring native vision capabilities and support for a context window of 1 million tokens. The model's architecture is based on the MoE mechanism, ensuring high efficiency in processing complex tasks.
The model demonstrates significant progress in code generation and agent operations. Token consumption has dropped by up to 50% compared to the Qwen3.8-Max version, leading to a reduction in the average cost per task execution by more than 60%.
Independent test results confirm the novelty's leadership: Agent Arena saw a 7.5% growth after processing 14,700 real-world tasks. In the Code Arena category, the WebDev model climbed to sixth place, improving its score by 74 points.
API access is provided via QwenCloud, Qwen Studio, QwenWork, and Qoder services. Pricing includes $1.5 per million input tokens, $4.5 per output token, and $0.075 for caching, making it 70% cheaper than the previous generation.
The development team also announced the imminent release of open weights for smaller models in the Qwen4 lineup. Special attention is given to the compact Qwen4-27B version, which can operate on systems with as little as 16 GB of memory.










