Alibaba Releases Qwen3.8‑Max, a 2.4 Trillion‑Parameter Multimodal AI

Overview

• Alibaba launched Qwen3.8‑Max on Monday as its largest model to date, saying it has 2.4 trillion parameters, can handle text, images and video, and supports up to 1 million tokens per request.
• The company says the model uses a mixture‑of‑experts architecture that activates roughly 95 billion parameters at a time to lower compute needs and shorten response times.
• Alibaba announced the model weights will be available for public download next week through Model Studio and Alibaba Cloud, enabling outside customization and fine‑tuning.
• Company benchmarks claim performance comparable to Anthropic’s Fable 5 and Arena.AI rankings place Qwen3.8‑Max at the top of Chinese text models and second globally for visual tasks, though independent peer‑reviewed tests are still limited.
• The launch pushed Alibaba shares higher and sharpens competition among Chinese AI vendors as firms pursue open weights, efficiency tweaks and lower prices in a market shaped by U.S. chip export controls and constrained hardware access.