Qwen

Alibaba’s Tongyi Qianwen large model series. The officially announced 2.4T open-source flagship Qwen3.8 is launching soon—covering language, multimodality, code, and math.

Qwen (Tongyi Qianwen) is an open-source large model series developed by Alibaba DAMO Academy and one of the most influential open-source model families in the Chinese-speaking world today. On July 19, 2026, the official X account announced that Qwen3.8 is coming soon with open-sourced weights and a parameter scale of 2.4T, officially positioned as “on par with leading frontier AI models, second only to Fable 5”—marking the first time an open-source model enters the 2T+ parameter era. The preview version, Qwen3.8-Max-Preview, has already debuted simultaneously on Alibaba’s Token Plan, Qoder, and QoderWork, ready for developers to call right away. Unlike overseas open-source models such as Llama and Mistral, Qwen was designed from day one with a strong emphasis on Chinese corpora and Chinese-language scenarios, while staying competitive on English, code, math, and multilingual tasks. The family spans Qwen (language models), Qwen-VL (vision-language), Qwen-Audio (audio), Qwen-Coder (code), Qwen-Math (math), and Qwen-Agent (agent framework), covering virtually every mainstream direction in large model applications.

Pricing is primarily free and open source. Qwen series models are released under Apache 2.0 or the Qwen License, permitting research, commercial use, and derivative development; Alibaba Cloud also offers API access through its Bailian/DashScope platforms, billed by token. Developers can run the models locally with Ollama, vLLM, Llama.cpp, or transformers; enterprises can get high-concurrency, low-latency commercial APIs through Alibaba Cloud.

In terms of ecosystem positioning, Qwen is the strongest Llama alternative in the Chinese market and a major contributor to the global open-source community. Its downloads and derivative model counts on Hugging Face have long ranked among the very top, and a vast number of Chinese AI applications, RAG systems, and Agent platforms are built on Qwen. If you need an open-source model mainline with excellent Chinese support, a rich range of model sizes, and a thriving community ecosystem, Qwen is the most reliable choice.

Note: Licenses vary slightly across Qwen versions—confirm the licensing terms of the specific model before commercial use. Ultra-large models (e.g., 110B) place extreme demands on local VRAM, so quantized versions or cloud APIs are recommended.

Model Release Timeline

Date	Version/Model	Highlights
2023-04	Tongyi Qianwen 1.0	Alibaba’s first large language model goes public
2023-11	Qwen1.5	Open-source release with multiple sizes
2024-06	Qwen2	Major gains in Chinese and English; expanded multilingual support
2024-09	Qwen2.5	Full series open-sourced; 128K context
2025-01	Qwen2.5-VL / Coder / Math	Specialized vertical-domain models
2025-05	Qwen3	Third-generation foundation model with enhanced reasoning and Agent capabilities
2026-07-19	Qwen3.8	Officially announced as the upcoming 2.4T-parameter open-source flagship; Preview live on Qoder/Token Plan; open weights coming soon
2026-07-21	Qwen-Image-3.0	Third-generation image generation foundation: 4.5k-token instruction input, 10px fine-text rendering, native support for 12 languages

Source: Agentopia · Full site-wide machine catalog at /llms.txt