Alibaba Unveils Qwen3.8-Max With 2.4 Trillion Parameters
Qwen3.8-Max pushes multimodal AI from understanding to continuous autonomous work, demonstrated by a 125-hour run that reproduced a research paper and proposed 18 improvements.
Reporting from 1 source: GIGAZINE.
Alibaba's Qwen team announced Qwen3.8-Max, a multimodal AI with 2.4 trillion total parameters and 95 billion active parameters. It handles text, images, and video, with a 1 million token context. In some benchmarks it surpassed Anthropic's Claude Fable 5. The model ran 16 days autonomously, producing 265 commits and 127 pull requests.
The Qwen team positioned Qwen3.8-Max as a coding and Cowork model, meaning it collaborates with humans on long-term tasks. Its Mixture-of-Experts design keeps inference costs down while preserving the scale of a 2.4 trillion parameter model.
In a demonstration of autonomy, the model worked on a repository for about 16 days, accumulating 265 commits, 127 pull requests, and 151 issues. In a separate experiment, it reproduced a research paper on LLM data selection within 37 hours, then spent 88 more hours testing 18 improvement proposals across four rounds.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.