Alibaba unveiled Qwen 3.8 Max on July 19 at Shanghai World AI Conference, a 2.4-trillion-parameter model claimed second only to Anthropic Fable 5.

The sparse Mixture-of-Experts multimodal model processes text, images, video, and documents with a one-million-token context window. Alibaba released it as preview on its Token Plan, Qoder, and QoderWork platforms, offering 10% discount pricing.

The Specs and Performance Claims

With 2.4 trillion parameters, Qwen 3.8 Max sits between frontier models and open-weight alternatives. Alibaba made bold positioning claims about performance, though preview launched without benchmarks or third-party validation.

The company said it plans open-weight release soon with no timeline given. This mirrors industry trends where large model makers release gated previews first.

Pricing and Developer Access

During preview, model runs on credit-based tiers: $6 Lite plan gets 2,500 credits weekly, $68 Pro gets 40,000 credits with concurrent agents. Developers access it through Alibaba platforms without upfront fees.

The preview launch positions Qwen to capture developers exploring alternatives to American and European AI models. Alibaba infrastructure advantage in China gives it leverage serving domestic demand.

The race to build frontier AI models now involves spending at scales previously reserved for semiconductor manufacturing alone.