Alibaba is going all in on Qwen…

Overview

Alibaba unveiled a $52 billion, three-phase roadmap toward artificial superintelligence (ASI) at the Apsara 2025 conference. Their ambitious plan includes groundbreaking AI models like Qwen 3 Max, Qwen 3VL, and Qwen 3 Omni, aiming to position Alibaba as the Android of the AI era. The roadmap outlines a progression from generalized AI understanding to autonomous action and self-iteration, with a target of achieving ASI by 2032.

Notable Quotes

- This is where we either all die or live in a gluttonous utopia, and Alibaba thinks we'll be there by 2032. – On the implications of ASI.

- Quen 3VL is now the tallest dwarf with 39% accuracy. – On the challenges of vision-language AI models.

🚀 Alibaba’s Three-Phase Roadmap to ASI

- Phase 1: Emergence of intelligence, where AI achieves generalized understanding by learning vast amounts of human knowledge. This phase represents the current state of AI development.

- Phase 2: Autonomous action, enabling AI to use human languages and tools to assist in tasks. This is where AI stands today.

- Phase 3: Self-iteration, where AI connects to the physical world and learns independently, potentially leading to ASI by 2032.

- The speaker expressed skepticism about the timeline but acknowledged Alibaba’s rapid progress.

🧠 Qwen 3 Max: A Trillion-Parameter Powerhouse

- Qwen 3 Max boasts over 1 trillion parameters and was pre-trained on 36 trillion tokens using a mixture-of-experts architecture.

- The model performs well on standard benchmarks, with a reasoning variant currently in training.

- The speaker humorously noted the model’s impressive capabilities despite being non-thinking.

📸 Qwen 3VL: Vision-Language Model Innovations

- Designed to process and understand visual inputs like images and videos, Qwen 3VL leads the new Clockbench benchmark with 39% accuracy.

- The model addresses a common challenge for AI: interpreting analog clocks.

- Bonus points for being open-source and open-weight, making it accessible for broader use.

🎙️ Qwen 3 Omni: Multimodal AI for All Senses

- Qwen 3 Omni integrates vision, hearing, reading, and talking capabilities, enabling multimodal interactions.

- The model opens possibilities for applications like high-quality AI companions.

- The speaker humorously cautioned against using the model for creating AI girlfriends.