AI/TLDR

Qwen3.8-Max-0902

The 2 September 2026 snapshot of Alibaba's Qwen3.8-Max — a post-training update for coding and agentic office work, at the same 1M context and $2 / $6 per million tokens.

Qwen-MaxAPI onlyGenerally available
Released
2 Sep 2026
Context
1M
Parameters
2.4 trillion (Mixture-of-Experts) — the Qwen3.8-Max base this snapshot updates
Input
$2.00 / 1M tokens
License
Proprietary (hosted API)
Coverage
1 story

Overview

Qwen3.8-Max-0902 is the 2 September 2026 snapshot of Alibaba's Qwen-Max flagship. QwenCloud describes it as an upgraded snapshot with enhanced coding for complex engineering projects, improved collaborative-agent performance with multi-tool orchestration, and refined native vision understanding across chart reasoning and document parsing. It is a post-training update, not a new base model.

The specifications carry over from Qwen3.8-Max: a 1M-token context window, up to 131K tokens of output (in both standard and thinking modes), and image, text and video input with text output. The underlying base is the 2.4-trillion-parameter sparse Mixture-of-Experts model documented on QwenCloud's Qwen3.8-Max page.

Pricing is unchanged from the August 2026 Qwen3.8-Max release: $2 per million input tokens, $6 per million output tokens, $0.25 per million tokens for implicit cache input, $2.50 for explicit cache creation and $0.17 for explicit cache reads.

The snapshot is served as model id `qwen3.8-max-0902`, with `qwen3.8-max-2026-09-02` documented as an alias. Alibaba Cloud's Model Studio documentation references the same id, so the snapshot is addressable from both QwenCloud and Alibaba Cloud Model Studio.

Released2026-09-02
LicenseProprietary (hosted API)
WeightsAPI only
Parameters2.4 trillion (Mixture-of-Experts) — the Qwen3.8-Max base this snapshot updates
Context1M
Max output131K tokens
ArchitectureA dated snapshot of Qwen3.8-Max rather than a new architecture: the QwenCloud model card describes the same sparse Mixture-of-Experts base, re-post-trained. Alibaba serves it under the id qwen3.8-max-0902, with qwen3.8-max-2026-09-02 as an alias.
ModalitiesText, Image, Video
StatusGenerally available

Pricing

Input$2.00 / 1M tokens
Cached input$0.25 / 1M tokens
Output$6.00 / 1M tokens

Implicit cache input is $0.25 per 1M tokens; explicit cache creation is $2.50 and explicit cache reads $0.17 per 1M tokens.

Pricing source ↗

Strengths

  • Targets coding on complex engineering projects and multi-tool agent orchestration, per QwenCloud's own description of the snapshot
  • Refined native vision understanding for chart reasoning and document parsing
  • 1M-token context with up to 131K tokens of output, including in thinking mode
  • Same price as the base Qwen3.8-Max — $2 / $6 per million tokens, with cached input at $0.25
  • Pinnable: qwen3.8-max-2026-09-02 addresses this exact snapshot rather than a floating alias

Best for

  • Reach for it for long-horizon coding agents that need a large context and cheap cached input
  • Reach for it for multi-tool agent workflows over office documents and long task chains
  • Reach for it when a workload mixes text with charts, screenshots or video that must be read natively
  • Reach for it when you need to pin a dated snapshot so model behaviour does not shift under a running system

How to access

ProviderModel ID
QwenCloud ↗qwen3.8-max-0902
Alibaba Cloud Model Studio ↗qwen3.8-max-0902

Qwen-Max — every version

The full lineage of the Qwen-Max line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.

VersionReleasedContextLicense
Qwen3.8-Max-0902current2026-09-021MProprietary
Qwen3.8-Max2026-08-031MProprietary (weights published as Qwen3.8-2.4T-A95B)
Qwen3.8-Max-Preview2026-07-19Proprietary
Qwen3.7-Max2026-05Proprietary
Qwen3-Max2025-09Proprietary
Qwen2.5-Max2025-01-29Proprietary

FAQ

What is Qwen3.8-Max-0902?

Qwen3.8-Max-0902 is the 2 September 2026 snapshot of Alibaba's Qwen3.8-Max flagship. QwenCloud describes it as an upgraded snapshot with enhanced coding for complex engineering projects, improved collaborative-agent performance with multi-tool orchestration, and refined native vision understanding for chart reasoning and document parsing.

How is it different from Qwen3.8-Max?

It is a post-training update on the same base rather than a new model. The context window (1M tokens), maximum output (131K tokens), modalities (image, text and video in, text out) and pricing all carry over from Qwen3.8-Max; what Alibaba changed is the training that sits on top.

What does it cost?

QwenCloud lists $2 per million input tokens and $6 per million output tokens, with implicit cache input at $0.25, explicit cache creation at $2.50 and explicit cache reads at $0.17 per million tokens — the same rates as the base Qwen3.8-Max.

Which model id should I call?

Use qwen3.8-max-0902, or the alias qwen3.8-max-2026-09-02 if you want to pin this exact dated snapshot. Both are documented on QwenCloud, and Alibaba Cloud Model Studio's text-generation documentation references the same id.

Are the weights open?

No. Qwen3.8-Max-0902 is a hosted, API-only snapshot. Alibaba has not published downloadable weights for it.