Alibaba Group's Qwen models are set to scale significantly, with plans for Qwen4, Qwen4.5, and Qwen5 to reach parameter counts between 5T and 10T, building on the 2.4T parameters of Qwen3.8. The company is also focusing on architectural innovations to reduce training and inference costs while increasing model size. Alibaba is exploring recursive self-improvement (RSI) to accelerate model development, with experiments showing Qwen participating in chip design and optimizing inference infrastructure. In multimodality, Alibaba previewed an "Agent Video" model for November, aiming for longer, more controllable, and intelligent video generation, alongside upgrades to image, audio, and music generation models. AI
IMPACT Alibaba's aggressive scaling and focus on AI-driven development could accelerate the pace of LLM advancement and adoption.
RANK_REASON Alibaba Group's Qwen models are frontier models, and the announcement details their scaling plans and new capabilities. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Alibaba Group
- Anthropic
- Cloud栖 Conference
- Dario Amodei
- Kimi k3
- Li Feifei
- Liu Dayiheng
- Morgan Stanley
- Qwen
- Qwen 2.5
- Qwen 3.8
- Qwen 3.8-Flash
- Qwen 4
- Qwen 4.5
- Qwen 5
- Qwen-Audio-3.1
- Qwen-Image-3.1
- T-Head
- Wu Yongming
- Zheng Bo
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
