Alibaba Unveils Full-Stack AI Roadmap From Chips to Agents
Article excerpt
Alibaba Group unveiled a full-stack artificial intelligence strategy and technology roadmap spanning chips, cloud infrastructure, models and agents. The announcements were made at Alibaba Cloud’s annual flagship technology event, the Apsara Conference. Key updates included Qwen foundation and multimodal models, a new proprietary AI chip, an agentic cloud designed for running AI agents, and an AI agent platform for smartphones. Alibaba said its next-generation Qwen 4 model is currently in training and also outlined a roadmap for the upcoming Qwen 4.5 and Qwen 5 model series. The future models are expected to scale to between 5 trillion and 10 trillion parameters. Alibaba also presented results from recursive self-improvement (RSI), in which models iteratively improve themselves based on feedback from experiments and validation. According to Alibaba, Qwen3.8-Max autonomously carried out an entire process spanning pipeline design, data validation, iterative experiments and error diagnosis over a month, completing 33 improvement cycles. Using autonomous learning optimization and post-training techniques, the updated Qwen3.8-Max increased its Intelligence Index score on AI model evaluation site Artificial Analysis (AA) from 40 to 45. Alibaba also said the model conducted more than 60 hours of self-improvement across the entire chip-design process in an experiment. It called...
Keep reading with a free account
The rest of this article, and every signal for Alibaba Cloud, is in your free account.
Extracted from this sentence
Alibaba also announced broader technology and service upgrades centered on its agentic cloud strategy, extending its efforts beyond models and processors.
