Alibaba's Qwen3.8-Max Pushes Open-Source AI to 2.4 Trillion Parameters

On Tuesday, Alibaba released Qwen3.8-Max, a new open-weight language model with about 2.4 trillion parameters. The launch marks the largest public Chinese model so far, narrowing the gap with the leading closed systems from the United States. The model uses a sparse mixture-of-experts design, which means only a portion of the weights are active for any single query.

Qwen3.8-Max supports a context window of one million tokens, allowing it to read long documents, full code repositories, or hours of meeting transcripts in a single pass. In public coding and reasoning benchmarks, the model ranks just behind Anthropic's top Claude family and ahead of most other open releases. The release notes also highlight stronger tool use and more stable long-context recall.

The weights are released under a permissive license, so companies can download, fine-tune, and run the model inside their own data centers. For Chinese enterprise customers, this is a practical answer to the question of how to deploy large AI without sending sensitive data to overseas APIs. The price of running the model through Alibaba's cloud has also been reduced by about thirty percent compared with the previous Qwen generation.

Industry watchers see the launch as a signal that open-source large models are catching up fast. With sparse architectures cutting the cost of inference, even trillion-parameter models can serve real users at a reasonable price. The race is no longer just about who has the biggest model, but who can deliver the best mix of capability, openness, and operating cost.