Qwen3.8-Max
Alibaba's first open Max-class model: 2.4T parameters, 95B active, 1M context, with text, image and video input.
In short. Qwen3.8-Max is the biggest and strongest model from Alibaba's Qwen team, and the first of its top tier whose weights anyone can download. It handles text, images and video and is built to finish long, multi-step jobs like coding and report writing. Running it yourself takes a multi-GPU data-center server.
Summary
Qwen3.8-Max is the flagship of Alibaba's Qwen3.8 generation. Alibaba announced it and opened the API on 2 August 2026, and published the weights as Qwen3.8-2.4T-A95B on 12 August 2026. It is the first time a Qwen "Max"-class model has been released with open weights.
It is a mixture-of-experts model with 2.4 trillion parameters in total and 95 billion active per token, built on the architecture of Qwen3.5. It takes text, images and video as input, writes text, and supports a context window of up to 1 million tokens. Alibaba aims it at coding, professional work, research and long multi-step agent tasks. On its own blog Alibaba reports a Terminal Bench 2.1 score of 86.6; these figures are self-reported and use harnesses chosen by Alibaba.
Two checkpoints are published: the original BF16 weights at about 4.9 TB and an FP8 version at about 2.5 TB. The license is the custom Qwen3.8-Max License: free for most uses, a name credit for products with over 100 million monthly users or US$20 million of monthly revenue, and a separate agreement for large hosting and AI work-assistant businesses above US$50 million in revenue.
Good for
- Self-hosting a frontier-class model for coding and long agent tasks
- Understanding long documents, charts and video in one pass
- Teams that want an alternative to closed API models and already operate large GPU clusters
- Long financial reports and PDFs, which Alibaba highlights (documents of more than 200 pages)
Watch out for
- Even the FP8 checkpoint is about 2.5 TB, so it needs a multi-node GPU setup
- Large hosting businesses and AI work-assistant products may need a separate license from Qwen
- Benchmark numbers are Alibaba's own and use its chosen harnesses
- For a single GPU, use Qwen3.8-27B instead
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Estimates for the 2.4T-A95B variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.
| Variants | Parameters | BF16 | FP8 / INT8 | 4-bit (Q4) | Fits on |
|---|---|---|---|---|---|
| 2.4T-A95B BF16 | 2.45T 95B active | 5.63 TB | 2.81 TB | 1.69 TB | Multi-node cluster |
| 2.4T-A95B FP8 FP8 | 2.45T 95B active | 5.63 TB | 2.81 TB | 1.69 TB | Multi-node cluster |
Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.
The license in plain words
An MIT-style license from Alibaba's Qwen team for its flagship model. Free for almost everyone, with a name-credit rule for huge products and a separate deal for large hosting and AI work-assistant businesses.
- Attribution or model name must be shown
- Separate deal needed to sell hosted access above a threshold
More about Qwen3.8-Max License
See also
- Kimi K3.1 Pro vs Qwen3.8-Max: which giant open model fits you?
- Best open-source LLMs in 2026
- Open licenses: free to build on, but hosting needs a deal
- Qwen3.8: three language models, three licenses, and a fourth for images
Frequently asked questions
- What is Qwen3.8-Max?
- Qwen3.8-Max is the biggest and strongest model from Alibaba's Qwen team, and the first of its top tier whose weights anyone can download. It handles text, images and video and is built to finish long, multi-step jobs like coding and report writing. Running it yourself takes a multi-GPU data-center server.
- Can I use Qwen3.8-Max commercially?
- Yes, with conditions. Qwen3.8-Max is released under the Qwen3.8-Max License, which allows commercial use but adds extra rules: Attribution or model name must be shown; Separate deal needed to sell hosted access above a threshold. Read the license text before you ship a product.
- Is Qwen3.8-Max open source?
- Qwen3.8-Max is open-weight, not open source in the strict sense. You can download the model files, but the Qwen3.8-Max License adds conditions that the Open Source Initiative definition does not allow.
- Can I run Qwen3.8-Max on my own computer?
- Not realistically. The smallest variant, 2.4T-A95B, needs about 1.69 TB at 4-bit precision, more than a single GPU server holds. It is meant for clusters or hosted services.
- When was Qwen3.8-Max released?
- Qwen3.8-Max was released by Alibaba Qwen on August 12, 2026.
- How long a context window does Qwen3.8-Max have?
- Qwen3.8-Max supports a context window of up to 1M tokens.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- Qwen3.8-Max: A New Bar for Coding and Cowork (Qwen blog, 2 Aug 2026) qwen.ai
- Qwen3.8 repository and release news (GitHub) github.com
- Alibaba Unveils Qwen3.8-27B and Releases Weights of Qwen3.8 Flagship Model (Alibaba Cloud, 17 Aug 2026) alibabacloud.com
- Qwen3.8-2.4T-A95B model card and license huggingface.co
Release date and facts last verified November 8, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| Kimi K3.1 Pro Moonshot AI | 2.8T 104B active | 1M | Kimi K3 License (Conditional) | 1.72 TB Multi-node cluster | |
| GLM-5.3 Z.ai (Zhipu AI) | 753B 40B active | 1M | GLM-5.3 License (Conditional) | 520 GB GPU server | |
| Mistral Large 4 Mistral AI | 1.05T 52B active | 1M | Mistral Open-Weight (Conditional) | 671 GB GPU server | |
| Qwen-Image-2.1 Alibaba Qwen | 7.1B | – | Qwen Research License (Non-commercial) | 20 GB Gaming GPU |