Qwen3.8-Max: 2.4 Trillion Parameters, Open Weights Next Week
Alibaba dropped Qwen3.8-Max on August 3, and this is the flagship card they had been holding back. 2.4 trillion parameters, MoE, a million tokens of context, natively multimodal. It ranks fifth on Text Arena, second on Vision Arena, fourth on the coding arena. Alibaba's stock rallied on the announcement, which tells you how much of China's AI story now rides on Qwen.
The detail that matters for agent people: Alibaba says Max can code autonomously for weeks. In one internal test it spent 16 days building and improving an AI coding tool on its own, writing, testing, fixing and refining with little human input. Every frontier lab is converging on this same pitch, but 16 days is among the longest-horizon claims anyone has put on paper so far.
The real move is distribution: model weights are scheduled for release next week. Alibaba has never open-sourced anything at this scale, and 2.4T parameters would make it by far the largest open-weights model ever. After Kimi K3's weights in July and DeepSeek V4 Flash on July 31, the Chinese labs are turning open weights into a coordinated strategy right as Washington debates restricting them. If the weights actually land next week, the "open frontier" stops being a slogan and becomes a 2.4-trillion-parameter fact.
API is live on Alibaba Cloud Model Studio. Announcement: https://www.alizila.com/alibaba-unveils-qwen3-8-max-most-capable-flagship-model-to-date/
← Back to all articles
The detail that matters for agent people: Alibaba says Max can code autonomously for weeks. In one internal test it spent 16 days building and improving an AI coding tool on its own, writing, testing, fixing and refining with little human input. Every frontier lab is converging on this same pitch, but 16 days is among the longest-horizon claims anyone has put on paper so far.
The real move is distribution: model weights are scheduled for release next week. Alibaba has never open-sourced anything at this scale, and 2.4T parameters would make it by far the largest open-weights model ever. After Kimi K3's weights in July and DeepSeek V4 Flash on July 31, the Chinese labs are turning open weights into a coordinated strategy right as Washington debates restricting them. If the weights actually land next week, the "open frontier" stops being a slogan and becomes a 2.4-trillion-parameter fact.
API is live on Alibaba Cloud Model Studio. Announcement: https://www.alizila.com/alibaba-unveils-qwen3-8-max-most-capable-flagship-model-to-date/
Comments