Qwen3.8-Max
Alibaba's largest model has 2.4T parameters (95B active) and 1M context, and is positioned for coding and 'cowork'. Open weights followed the next week.
Scales Qwen's agentic MoE to 2.4T total parameters. Announced alongside QwenWork. Caixin says it is the second Chinese model above 2T parameters (after Moonshot's Kimi K3) and ranked fourth on a web-dev coding leaderboard behind Claude Opus 5 models.
- Date
- Sunday, 2 August 2026
- Lab
- Alibaba (Qwen)
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Parameters | 2.4T total / 95B active Hugging Face card for the open Qwen3.8-2.4T-A95B: Terminal Bench 2.1 86.6, SWE-bench Pro 67.7 | company |
Blog 'Qwen3.8-Max: A New Bar for Coding and Cowork' dated 2026-08-02 per Releasebot; Caixin reports the launch on 2026-08-03/04. Hosted Max adds multimodal and extended-context support relative to the open card. Ranking and Kimi K3 comparison are Caixin's claims. Alibaba shares rose about 7% in Hong Kong (Caixin).
Sources
- www.caixinglobal.com/2026-08-04/alibaba-releases-new-qwen-model-consolidates-ai-office-too
- releasebot.io/updates/qwen
- huggingface.co/Qwen/Qwen3.8-2.4T-A95B
This record was checked against its sources on 6 October 2026. How we check