Alibaba’s Qwen3.8-Max Raises the Bar with 2.4 Trillion Parameters

Alibaba has launched Qwen3.8-Max, a massive new AI model with 2.4 trillion parameters. This makes it one of the largest mixture-of-experts (MoE) models available today. The model handles not just text, but images and video as input, returning detailed text responses.

Qwen3.8-Max supports a context window of up to 1 million tokens. That means it can process about 750,000 words or their equivalent in one go. This size allows it to understand and analyze extremely long documents, TV series, or even 100-hour livestreams. It can turn this data into searchable knowledge bases.

The model is now generally available through a hosted API. Companies of any size can deploy it today. Pricing is set at $2.00 per million input tokens, $6.00 per million output tokens, and $0.25 per million cached input tokens. Open weights for both Qwen3.8-Max and a smaller 27 billion parameter version will ship next week.

The 27 billion parameter checkpoint is intended for on-premise deployments. It offers a more realistic path for companies wanting to run the model internally. However, the flagship 2.4 trillion parameter model remains cloud-hosted for now.

Strong Performance and Versatile Capabilities

Alibaba shared benchmark results showing Qwen3.8-Max performs very well. On Terminal-Bench 2.1, it scored 86.6, beating Anthropic’s Claude Opus 4.8 and Claude Fable 5. It only trailed behind GPT-5.6 Sol (max). On coding benchmarks like SWE-bench Pro and FrontierSWE, it scored 67.7 and 73.5 respectively, showing a major jump from its predecessor, Qwen3.7-Max.

The model also shines in vision tasks. It topped scores on OSWorld-Verified (86.1), Parametric CAD Bench (91.5), and OmniDocBench (92.1). These scores confirm its strength in processing complex visual data alongside text. Alibaba claims Qwen3.8-Max can recreate software apps from screenshots, generate interactive games and educational animations, and convert 2D floor plans into 3D models.

Autonomous Coding and Business Impact

One impressive test showed Qwen3.8-Max working autonomously for 16 days. It built and improved an AI coding tool on its own by writing, testing, and debugging code with little human input. This level of autonomy could revolutionize software development and automation.

The model supports use cases in coding, legal and financial document review, research, and architectural modeling. It integrates with OpenAI and DashScope platforms, making it compatible with existing AI workflows.

Alibaba’s Hong Kong-listed shares jumped 7% on the day of the release. Its stock in New York also rose 4.5% the following Monday. This shows strong investor confidence in the new model’s potential.

Alongside the model launch, Alibaba rolled out QwenWork, a workplace AI agent platform. This all-in-one productivity tool targets both individuals and businesses. It entered public beta on the same day, signaling Alibaba’s push to embed AI deeper into daily work.

Open-Weight Strategy and Global Competition

Qwen3.8-Max marks Alibaba’s return to open-weight model releases. After briefly focusing on proprietary models earlier this year, Alibaba is now releasing weights openly again. This shift highlights open-weight releases as a key point of competition within China’s AI industry.

Many top Chinese AI models, including Moonshot’s 2.8 trillion parameter Kimi K3, are open-weight. Alibaba’s move narrows the gap with US AI labs and increases competition with companies there. The model’s performance broadly matches or exceeds Anthropic’s Fable 5 on benchmarks.

On the Arena text model leaderboard, Qwen3.8-Max trails only a few models in Anthropic’s Opus family and Fable 5. It is beaten only by two Claude Opus models and Kimi K3 for frontend coding, and only by Fable 5 in visual analysis tasks. This shows it ranks near the top globally.

While parameter count is often seen as a quick measure of model power, more parameters don’t always mean better results. Still, Alibaba’s 2.4 trillion parameter Qwen3.8-Max shows strong capabilities and versatility across many AI tasks.

With this release, Alibaba sets a new bar for multimodal AI models that can handle huge context windows and diverse data types. It also pushes the AI industry forward by offering open access to one of the most advanced models available today.

Stay connected via Google News

Follow us for the latest AI News