Alibaba Cloud’s Agentic Cloud unfolds in three layers: AI Native Cloud for model training and inference, Agent Native Cloud for agent deployment and security, and the Context Engine for real-time data handling and long-term memory.

Qwen 4 is in active training, with plans to advance to Qwen 4.5 and Qwen 5, aiming for a scale of 5–10 trillion parameters and improvements in RSI-driven training efficiency and chip design.

Qwen Intelligence debuts as a business-facing, full-stack agentic solution for smartphones, enabling Qwen-powered agents for next-generation AI phones and cross-app task handling.

T-Head launches the Zhenwu V900 AI training and inference processor, delivering roughly three times the performance of its predecessor, with 216 GB GPU memory and 1,200 GB/s inter-chip bandwidth, targeting mass production in early 2027, alongside upgraded supernode servers and new CPUs (Yitian 720/730) for agentic AI tasks.

Alibaba Token Hub and Happyworld Arena introduce standardized world model evaluation across video generation, embodied AI, and spatial reconstruction, establishing benchmarks for model performance.

Multimodal model upgrades include Qwen3.8-LiveTranslate for real-time translation with lower latency, Qwen-Audio-3.1-TTS-Next for cinematic soundscapes, and Qwen-Audio-3.1-ASR/TTS/Realtime, with Qwen-Image 3.1 to accelerate creative design and e-commerce tasks.

Key AI Native Cloud components include the Platform for AI (PAI) with post-training optimizations, a high-throughput Cloud Parallel File Storage system, and HPN 8.0 Pro networking delivering 100-petabit bandwidth with low latency.

Alibaba unveils a comprehensive, full-stack AI roadmap spanning chips, cloud infrastructure, models, and agent-based tech, announced at the Apsara Conference in Hangzhou on September 22, 2026.

Context Engine features include Agent Context for enterprise memory and real-time knowledge linking, and OpenLake, a unified multi-modality data lakehouse that cuts costs by 38% and speeds queries by 40%.

Agent Native Cloud comprises AgentCore for building and running AI agents with lifecycle management and security, plus the Agent Security Center ensuring end-to-end security and compliance.