Understand this story
On November 3, 2026, Alibaba Cloud's Tongyi Qianwen launched the full Qwen3.8 model lineup, forming a complete product matrix. According to an article published by Alibaba Cloud Developer Community AI, the series adopts a sparse MoE architecture, with the entire lineup natively supporting multimodal input of text, images, and short videos, supporting a context window of up to one million tokens, and being compatible with the OpenAI API protocol. The flagship version Qwen3.8-Max has 2.4 trillion total parameters and activates 95 billion effective parameters during inference, with a maximum output of 131,072 tokens; after enabling deep thinking, the chain of thought can reach up to 262,144 tokens. There is also Qwen3.8-Flash, aimed at high-concurrency, low-latency scenarios. The article also provides cURL and Python streaming call examples, the billing methods for context caching and pay-as-you-go, and selection recommendations by scenario and deployment form. No further developments have been disclosed at present.
AI-generated from public reports
Developments
1 confirmed developments
Alibaba Cloud's Tongyi Qianwen launches the complete Qwen3.8 product matrix
Source:English text is being prepared.
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 3
Alibaba Cloud Developer Community AI
Alibaba Cloud Tongyi Qianwen Releases the Full Qwen3.8 Model Lineup Along with API Integration and Selection Guide
Alibaba Cloud Tongyi Qianwen has launched the full Qwen3.8 model lineup, which adopts a sparse MoE architecture, natively supports multimodal input of text, images, and short videos across the entire lineup, supports a context window of up to one million tokens, and is compatible with the OpenAI API protocol. The flagship version Qwen3.8-Max has 2.4 trillion total parameters, activates 95 billion effective parameters during inference, and has a maximum output of 131,072 tokens; when deep thinking is enabled, the chain of thought can reach up to 262,144 tokens. There is also Qwen3.8-Flash, aimed at high-concurrency, low-latency scenarios. The article also provides cURL and Python streaming call examples, the billing methods for context caching and pay-as-you-go, and selection recommendations by scenario and deployment form.