Alibaba launches closed-weight Qwen4-Max with 2.4 trillion parameters
The first Qwen4 model keeps 71 billion parameters active per token, reads a one-million-token window, and starts at $1.50 per million input tokens on QwenCloud. Alibaba says it uses up to ~50% fewer tokens than Qwen3.8-Max and plans open weights only for smaller Qwen4 models.
- Qwen4-Max is on QwenCloud, QwenWork, Qoder, QwenStudio and Vercel AI Gateway under the name alibaba/qwen4-max.
- A further 328 billion parameters are used for N-gram Embedding, and the model accepts images as well as text.
- On Alibaba's Work360 test the score was 52.69, against 47.74 for Qwen3.8-Max.