Alibaba's Qwen3.8-Max launches with 2.4T parameters
Alibaba's Qwen3.8-Max hits 2.4T parameters with a 1M token window...
Alibaba’s Qwen team launched **Qwen3.8-Max** on August 3, 2026, as its flagship model for 2026—a **2.4 trillion parameter Mixture-of-Experts (MoE) model** with **95 billion active parameters per query**. The model supports a **1 million token context window** (991K input/131K output tokens) and natively handles text, images, and video, positioning it for complex, multi-step tasks like repository-scale coding, long-document analysis, and 3D visualization from 2D floor plans. Weights for Qwen3.8-Max and a smaller **Qwen3.8-27B** checkpoint will be released open-source next week on Hugging Face and ModelScope, marking Alibaba’s return to open-sourcing top-tier models after keeping several 2026 releases proprietary.
The model debuts with competitive benchmarks, scoring **86.6 on Terminal-Bench 2.1** (ahead of Claude Opus 4.8 but behind GPT-5.6 Sol) and **93.0 on PaperBench**, while excelling in multimodal tasks like **OSWorld-Verified (86.1)** and **OmniDocBench 1.5 (92.1)**. Pricing starts at **$2.00 per million input tokens and $6.00 per million output tokens**, with rate limits of **2M tokens/minute** and **15K requests/minute**. Alibaba positions Qwen3.8-Max for agentic workflows, citing improvements over its predecessor in tasks like **DeepSWE 1.1 (56.6 vs. 21.6)** and **JobBench (53.4 vs. 31.3)**—though vendor-reported numbers await third-party verification.
- Qwen3.8-Max is a 2.4T parameter MoE model with 95B active parameters per query, supporting 1M token context windows and multimodal inputs (text/image/video).
- Priced at $2M/$6M per input/output token, it outperforms some competitors in benchmarks like Terminal-Bench 2.1 (86.6) and PaperBench (93.0), with open weights releasing next week.
- Designed for long-horizon agentic tasks—coding, research, and 3D visualization—with 2M token/minute rate limits and five built-in tools (e.g., code_interpreter, web_search).
Why It Matters
First open-weights Qwen-Max model fuels competition in trillion-parameter AI, enabling advanced agentic workflows for enterprises and developers.