Alibaba unveils plan: Qwen 4.5 and Qwen 5 to scale to 5

Alibaba unveils plan: Qwen 4.5 and Qwen 5 to scale to 5–10 trillion parameters

At the Apsara Conference in Hangzhou, Alibaba announced that Qwen 4 is already in training, that future Qwen 4.5 and Qwen 5 models will scale to 5–10 trillion parameters compared to 2.4 trillion in the flagship Qwen 3.8 Max, and unveiled the Zhenwu V900 AI accelerator, called China's most powerful AI chip.

Compiled by Sergey KostenchukPublished 2026-09-22Updated 2026-09-22

2026-09-22 Business NVIDIA

🚨 Alibaba: Qwen 4.5 and Qwen 5 to scale to 10 trillion parameters

At the Apsara Conference in Hangzhou, Alibaba announced: Qwen 4 is already in training, and Qwen 4.5 and Qwen 5 will scale to 5–10 trillion parameters compared to 2.4 trillion in Qwen 3.8 Max. The Zhenwu V900 accelerator was unveiled — China's top chip.

🌍 Alibaba is building an AI stack without NVIDIA: T-Head accelerators (V900 is three times faster than M890) with clusters of up to 500,000 chips and cloud capacity up to 20 GW by 2032. Recursive self-improvement was announced, but parameters alone do not guarantee quality.

👤 If the plan is realized, Qwen models will become significantly smarter in agentic scenarios, code, and multi-step tasks, and competition will lower inference prices. The benchmark is V900 benchmarks and the Qwen 4 release, not promises.

Source 1: https://www.reuters.com/business/retail-consumer/alibaba-plans-ai-model-with-5-trillion-10-trillion-parameters-unveils-new-chip-2026-09-22/ Source 2: https://finance.yahoo.com/technology/ai/articles/alibaba-plans-ai-model-5-025213560.html

Sources