From the source
Alibaba is introducing Qwen4-Max, a 2.4-trillion-parameter model built on the new Qwen4 architecture with native vision and a one-million-token context window.
- The company says Qwen4-Max is a significant step up from Qwen3.8-Max, delivering stronger performance across nearly every capability it evaluated.
- Alibaba reports that Qwen4-Max outperforms Qwen3.8-Max with fewer tokens: on Work360, token usage drops by approximately 24% and 35% at Xhigh and High effort, and on Terminal-Bench 4.0 by approximately 48% and 49%.
- Alibaba has also lowered API prices compared with Qwen3.8-Max, with particularly large reductions in cached-input pricing; in its Work360 evaluation, the average cost per task drops from $4.43 to $1.67, a 62% reduction.
- The model is built on a 2.4T backbone with 71B activated per token and an additional 328B N-gram Embedding parameters, using GDN plus Qwen Sparse Attention, Gated Residual branches and N-gram Embedding.
- Alibaba says pre-training at that scale progressed smoothly, with loss decreasing steadily and no loss spikes, and that Qwen4-Max-Base improves on Qwen3.8-Max-Base in every domain with approximately 5–13% lower BPB while activating 25% fewer parameters.
- On the Work360 benchmark, Qwen4-Max scored 52.69, up from 47.74 for Qwen3.8-Max.

