88Z AI, 13 August — The Qwen team has open-sourced the Qwen3.8-2.4T-A95B model, marking the first time weights for a Qwen-Max level model have been released. This MoE model features 2.4 trillion total parameters with 95 billion activated per token and natively supports a 256K context window. It focuses on enhancing capabilities in coding, office tasks, scientific research, and long-cycle agents, leading in benchmarks like PaperBench and TerminalBench. The model supports deployment via frameworks such as SGLang and
At a glance
- First open-source release of Qwen-Max level weights, featuring 2.4T total parameters and 95B activated parameters.
- Natively supports 256K context, enhances coding and research capabilities, and supports deployment via SGLang/vLLM.
The update
The Qwen team has officially open-sourced the weights for the Qwen3.8-2.4T-A95B model, marking the first public release of weights for a Qwen-Max level model. As a Mixture-of-Experts (MoE) architecture, it boasts 2.4 trillion total parameters, activates 95 billion parameters per token, and natively supports an ultra-long context window of 256K. The model focuses on enhancing performance in programming, office assistance, scientific research, and long-cycle agent tasks. It achieves leading results in authoritative benchmarks such as PaperBench and TerminalBench. Furthermore, the model is compatible with mainstream deployment frameworks like SGLang and vLLM, features thinking mode enabled by default, and allows users to flexibly adjust reasoning depth according to their needs.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 魔搭ModelScope社区Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.