Alibaba releases the 2.4 trillion-parameter Qwen3.8 as a free open model

Summary

Alibaba's Qwen research team released the 2.4 trillion parameter model Qwen3.8-2.4T-A95B as a free open-source offering, bringing this model to the same level as OpenAI and Anthropic's top-tier models.

Why it matters

This release has the potential to democratize access to a massive Chinese AI model, shifting the balance of global AI development. Benchmark scores suggest that the model can match or surpass Western leading models, and its extended context window (up to 1 million tokens) opens new possibilities for long-text reasoning and multimodal tasks.

Highlights

• A total of 24 trillion parameters, with 950 billion active; requires data-center scale resources.
• Benchmark tests show that Qwen3.8 Max surpasses OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5 models on several metrics.
• Native support for visual input, chain-of-thought mode, and a default 1 million token context window.

Critical take

Although the model is open source, its massive size may limit practical use to organizations with large computational resources, and performance claims rely heavily on internal benchmarks rather than independent peer-reviewed evaluations.