Skip to content
Alibaba Cloud

Qwen3.8-Max

Alibaba's largest model, built for coding, office work and long-running agents, at a mid-range price.

Quality

Modality

multimodal

Context

1M tokens

Access

open weights

Fabian's Take

FM

"I haven't used Qwen myself recently, but I've seen some of the things people built with it. From those results, I'd put it around or slightly above the mid-tier models of the frontier labs. That's a serious model, just not one I'd reach for first."

Qwen3.8-Max is the flagship of Alibaba’s Qwen3.8 series, announced on 3 August 2026, with an updated snapshot from 2 September. It’s a mixture-of-experts model with about 2.4 trillion parameters, of which 95 billion are active for each token, and Alibaba aims it at coding, office tasks and agents that work through long jobs on their own.

Where it sits

It replaced Qwen3.7-Max, which was API-only. With 3.8, Alibaba published the weights, under its own license rather than Apache 2.0. In practice that changes little for most readers: at this size you reach it through Alibaba Cloud’s Model Studio or OpenRouter, not on your own hardware.

At $2 per million input tokens and $6 per million output tokens, it’s cheaper than the frontier labs’ top models (Claude Opus 5.5 costs $4 and $20) and several times the price of the cheapest workhorses like DeepSeek V4.1 Flash.

The Verdict

Best for: Coding, multi-tool agents and office tasks at mid-tier prices.

Pros

  • Judging by what others have built with it: around or slightly above the frontier labs' mid-tier models
  • 1M-token context, and reads images and video
  • Open weights, so it can be hosted outside Alibaba's cloud

Cons

  • 2.4 trillion parameters: hosting it yourself is a data center job
  • Published under Alibaba's own license, not Apache 2.0; read the terms before commercial use
  • Mostly reached through Alibaba Cloud or OpenRouter, not through the everyday apps

Specs

  • Pricing $2/M input, $6/M output (Singapore and US deployments) · $1.65/$4.95 on Alibaba's global deployment
  • Cost Tier moderate
  • ⚡
    Speed Tier moderate
  • License Qwen3.8-Max License (custom)
Developer Docs