At its Apsara Conference in Hangzhou on 2026-09-22, Alibaba laid out its model roadmap in a keynote by Group CEO Eddie Wu. The company said its next-generation model, Qwen 4, “is currently in training,” and announced a roadmap for Qwen 4.5 and Qwen 5 “projected to scale up to 5 to 10 trillion parameters.” Wu also set a goal of scaling Alibaba’s data center capacity to more than 20 gigawatts globally by 2032.
The most notable claim concerned what Alibaba calls recursive self-improvement (RSI) driven by empirical feedback. According to the release, over “a month of fully automated runs - spanning pipeline design, data validation, iterative experimentation, and error diagnosis - Qwen3.8-Max completed 33 iterative cycles,” and through autonomous training optimisation and post-training the updated model raised its Artificial Analysis score from 40 to 45. In a separate chip-design experiment, Alibaba says the model made more than 10,000 EDA tool calls over 60-plus hours and reduced chip area by 42 percent with no loss of performance.
On the product side, Alibaba introduced AgentCore, an enterprise platform to build, run, govern and monitor AI agents across their lifecycle, and Qwen Intelligence, a Qwen-powered agent platform offered to phone makers for cross-app tasks on smartphones. It also named new multimodal models including Qwen3.8-LiveTranslate, which it says cut latency from 2.8 to 2.3 seconds, and updated Qwen audio models.
Why it matters: a frontier lab publicly claiming a month-long, fully automated loop that measurably improved its own flagship model is a concrete marker in the move from AI-assisted to AI-driven model development, and the 5 to 10 trillion parameter target signals that Chinese labs still plan to scale pretraining. What it does not show: the release gives no technical report, no detail on how much human oversight the “fully automated” runs had, no definition of which Artificial Analysis measure moved from 40 to 45, and the chip-area result has no independent verification. Qwen 4 itself was not released.