Back to home
AlibabaReleased 2026-08-03

Qwen3.8-Max

Qwen3.8-Max is Alibaba's flagship released August 3, 2026. 2.4 trillion total parameters with 95B activated. LMArena score 1496 — global #5, the highest among Chinese models. $2/$6 per 1M tokens. Open weights coming soon.

Context
1M tokens
Input
$2/1M
Output
$6/1M
Modalities
textimage

Best For

Chinese language tasksLargest open-weights Chinese modelProduction workloads

Strengths

  • 2.4T params (largest Chinese model)
  • LMArena #5 globally, highest Chinese score
  • Pricing parity with GPT-5.6 standard but better Chinese
  • API: ¥12/¥36 per 1M domestic, $2/$6 international
  • Implicit cache: ¥1.5/$0.25 per 1M

Weaknesses

  • Weights not yet open (coming late August 2026)
  • New model — less ecosystem support
  • Qwen3.8-27B sync open (smaller variant)
  • Alibaba's stock +7% on launch

Related Alternatives

GPT-5.6

Better English, same price

GLM-5.2

Already open-source, MIT license

Frequently Asked Questions

How much does Qwen3.8-Max cost?

Qwen3.8-Max API costs ¥12 per 1M input and ¥36 per 1M output in China, or $2/$6 internationally. Implicit cache pricing is ¥1.5/$0.25 per 1M. Weights will open late August 2026.

Is Qwen3.8-Max better than GPT-5.6?

On Chinese tasks, yes — Qwen3.8-Max leads all Chinese models on LMArena with score 1496 (global #5, Chinese #1). On English and general reasoning, GPT-5.6 still leads slightly. For Chinese production workloads, Qwen3.8-Max is the best value.

See Qwen3.8-Max's full record in our database

Live pricing for all Alibaba plans, token cost calculator, and the latest price changes.

View Full Comparison