SURVEY RECORD N° 15 FILED 2026-08-13
← RETURN TO THE REGISTER

Qwen3.8-Max

~2.4T MoE with multimodal input; multilingual and vision breadth are its edge.

Context1M
Passage, per 1M in / out$2 / $6
AA Intelligencenot yet surveyed
Licenseopen
ModalitiesText, image, video in
Official siteqwen.ai →

In its favor

  • Open-license ~2.4T MoE with multimodal input
  • Multilingual and vision breadth are its edge
  • 1M-token context window
  • Backed by Alibaba's serving scale

Against it

  • Benchmark figures are vendor-reported, not independent
  • Priced well above DeepSeek at $2 / $6
  • GA rate card supersedes the July preview discount
  • Large to self-host at full size
open-weightmultimodalmultilingual1M contextvisionvendor-benchmarks

ASSESSMENT

Qwen3.8-Max is Alibaba's open-license flagship — a mixture-of-experts around 2.4 trillion parameters with multimodal input, and its edge is breadth: strong multilingual coverage and vision, backed by Alibaba's serving scale. It carries a 1M-token context and accepts text, image, and video in.

The standing comes with caveats. Benchmark figures are vendor-reported, not independent, and at $2 / $6 per million tokens it is priced well above DeepSeek's open models — the GA rate card having superseded the more generous July preview discount. At full size it is also large to self-host.

For multilingual and multimodal applications where an open license matters, Qwen3.8-Max offers a breadth few open models match. Teams optimizing purely for open-weight cost or independently verified capability will still look first to the DeepSeek tier.

FIELD NOTE — Priced well above DeepSeek; GA rate card supersedes July preview discount.

VISIT OFFICIAL SITE →