Qwen team releases Qwen4-Max, a 2.4-trillion-parameter model
Original title: Qwen4-Max
It adds native vision, a one-million-token window, and 71B active parameters, with lower API prices than Qwen3.8-Max.
Qwen Team introduced Qwen4-Max on a new sparse mixture-of-experts architecture for work, creative, and professional tasks. The model uses a 2.4-trillion-parameter backbone, activates 71 billion parameters per token, adds 328 billion N-gram embedding parameters, and includes native vision plus a one-million-token context. The team reports gains over Qwen3.8-Max on Work360, lower token use, and cheaper API pricing, and says pre-training ran without loss spikes.
Unconfirmed: Benchmark gains and cost figures are reported by Qwen Team and are not independently verified.
- Work360 score rose from 47.74 for Qwen3.8-Max to 52.69.
- Average Work360 cost per task fell from $4.43 to $1.67.
- The base model reports about 5–13% lower bits per byte than Qwen3.8-Max-Base.
- Architecture uses gated DeltaNet, sparse attention, and gated residual branches.
- Organizations
- Qwen Team
- AI models
- Qwen4-MaxQwen3.8-Max
From the original article
12 min read at qwen.ai2026/11/16 · 16 minute · 3115 words · QwenTeam丨Translations:简体中文 Today, we are introducing Qwen4-Max, your new AI teammate for work, creativity, and discovery. Qwen4-Max is built on the new Qwen4 architecture, with a 2. 4-trillion-parameter backbone, native vision capabilities, and a one-million-token context window.
By ilreb · qwen.ai
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.