Alibaba’s Qwen unveils 2.4 trillion parameter AI model, open weights coming next week

Alibaba’s Qwen team just dropped its biggest model yet. Qwen3.8-Max packs 2.4 trillion parameters and, according to internal benchmarks, sits right behind Anthropic’s Claude Fable 5 in performance rankings. The open weights are slated for release next week, which would make it one of the largest openly available AI models ever.

The model is multimodal, meaning it handles text, images, and potentially other data types rather than being limited to a single input format. Early access is already live through platforms like Token Plan and Qoder, with promotional discounts reaching up to 90% off standard pricing.

The timing is deliberate. Qwen3.8-Max arrived roughly one week after Moonshot AI launched its Kimi K3 model. What’s notably absent from the launch is a detailed breakdown of active parameters or comprehensive third-party benchmark results. Claiming 2.4 trillion total parameters is impressive on paper, but in mixture-of-experts architectures, only a fraction of those parameters activate for any given query. Without that number, it’s hard to compare apples to apples with Western counterparts.

There’s also the pricing pressure angle. Alibaba offering 90% discounts on API access isn’t just aggressive, it’s potentially destructive to the revenue models of proprietary AI providers. If the open-weight version performs anywhere close to the hosted version, developers building crypto applications powered by AI inference suddenly have a much cheaper option.

Alibaba’s Qwen series has been steadily building its open-weight portfolio since 2023, releasing progressively larger and more capable models. For Anthropic, the claim that Qwen3.8-Max trails only Claude Fable 5 should be both flattering and concerning. Flattering because it positions Claude at the top. Concerning because an open-weight model nipping at its heels undermines the premium pricing that funds Anthropic’s safety research.

Investors watching the AI-crypto intersection should track three things in the coming weeks: the actual benchmark results once independent testers get their hands on the open weights, the licensing terms that Alibaba attaches to commercial use, and whether decentralized compute networks see a measurable uptick in demand as developers rush to deploy the model outside centralized cloud providers.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.