Alibaba releases Qwen4-Max with 2.4T parameters and 1M-token context
Alibaba’s Qwen team has released Qwen4-Max, the first model in the Qwen4 generation. It is a proprietary sparse mixture-of-experts model with a 2.4T-parameter backbone and 71B parameters activated per token. The multimodal system accepts text, images and video, supports a 1M-token context window, and can produce up to 393K output tokens. It is available only through hosted services, including QwenCloud API, Token Plan, Qwen Studio, QwenWork and Qoder; Alibaba has not released weights for self-hosting.
The architecture adds Qwen Sparse Attention alongside Gated DeltaNet layers, gated residual connections and 328B N-gram Embedding parameters. Alibaba says the model maintained stable training at this scale and that Qwen4-Max-Base achieved 5–13% lower bits per byte than Qwen3.8-Max-Base while activating 25% fewer parameters.
On Alibaba’s comparison chart, Qwen4-Max ranked first on four of 16 benchmarks, including HLAB-Vals-1.1 at 21.3. It was weaker on some coding and vision tests, such as ProgramBench, where it scored 31.5. On Work360, its score was 52.69, compared with 47.74 for Qwen3.8-Max. Alibaba reports a 62% reduction in average Work360 cost per task, to $1.67. API pricing is $1.50 per million input tokens and $4.50 per million output tokens, with cached input at $0.075.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.