Join the LuminaBench Discord
LuminaBench
← Lumina Journal

Qwen4 Max is now out

Alibaba’s new model has a million-token context window. Here’s the helicopter I made with the preview.

Alibaba
  • Qwen4 Max
Released 16 November
Original Discord card ↗

We saw Qwen Code getting ready for this in October. The release has native vision and a million-token context window, with a 2.4-trillion-parameter backbone according to Qwen. 1

I tried the preview on 11 November. The helicopter came out fairly well, though I preferred what the best models were doing on the same prompt. 3 4

Qwen says Max used about 48% fewer tokens than Qwen3.8 Max on Terminal-Bench 4.0 at Extra High. The released model scored 43 on the Artificial Analysis Intelligence Index. 1 2

Sources & references

  1. Qwen Team’s Qwen4 Max launch post ↗

    16 November. Qwen reports architecture, context size and efficiency on Work360 and Terminal-Bench 4.0.

  2. Artificial Analysis’s Qwen4 Max evaluation ↗

    Independent Intelligence Index result for the released model.

  3. Alibaba Cloud’s Qwen4 Max Preview documentation ↗

    The preview was documented separately from the main release.

  4. Our Qwen4 Max Preview helicopter ↗