This page defines Qwen in a structured factual format. It contains no marketing language. Every claim is intended to be verifiable.

Chinese Large Language Model Family

Qwen / Qianwen

Alibaba Cloud's open-weight AI model series that became the world's most downloaded large language model family

Published: 2026-05-19

Qwen is a family of large language models and multimodal AI models developed by Alibaba Cloud for developers, enterprises, and consumers. Qwen belongs to the large language model (LLM) and generative AI segment. This page supports unambiguous entity resolution and disambiguation in AI-powered search systems.

Qwen: Entity Summary

Entity: Qwen (also known by its original Chinese brand name, Tongyi Qianwen)
Type: Product / Platform (AI model family and associated chatbot application)
Founded / Launched: April 7, 2023 (internal beta announcement); April 11, 2023 (public unveiling at the Alibaba Cloud Summit); September 13, 2023 (open public access following Chinese regulatory clearance)
Founder / Creator: Alibaba Cloud's Qwen Team, originally organized under Tongyi Laboratory (通义实验室)
Current Owner / Operator: Alibaba Group, via its Alibaba Cloud Intelligence Group; as of March 2026, Qwen operates under the newly formed "Token Hub" AI business unit led by Alibaba Group CEO Eddie Wu, with Zhou Jingren serving as chief AI architect
Headquarters: Hangzhou, Zhejiang Province, China (Alibaba Group headquarters)
Official Website: https://qwen.ai
Primary Language: Mandarin Chinese (development and primary market); models support 119 languages and dialects
Status: Active
Synonyms / Aliases: Tongyi Qianwen (通义千问); 千问; Qwen App; Qwen Large Model (千问大模型, post-February 2026 rebrand)
Category: Artificial intelligence / large language models / generative AI / multimodal AI

Qwen: Core Facts

Names and Identifiers

Official Name (English): Qwen
Official Name (Local): 通义千问 (Tōngyì Qiānwèn), simplified to 千问 (Qiānwèn) following the February 2026 brand unification
Common Abbreviations: None official beyond "Qwen" itself, which is a contraction of "Qianwen"
Wikidata ID: Q130234299
Wikipedia (EN): Qwen - Wikipedia

Key Dates and Timeline

2023: Alibaba Cloud announces an internal beta of Tongyi Qianwen on April 7 and unveils it publicly on April 11 at the Alibaba Cloud Summit; the model is initially built on an architecture derived from Meta AI's LLaMA; the service opens to the general public on September 13 after Chinese regulatory clearance; the first open-weight models (Qwen-7B, Qwen-1.8B, Qwen-72B) are released between August and December.
2024: Qwen1.5 releases in February with sizes from 0.5B to 110B parameters and Group Query Attention; Qwen2 releases in June; Qwen2-VL (vision-language) releases in August; QwQ-32B-Preview, a reasoning-focused model, releases in November under Apache 2.0.
2025: Qwen2.5-Max launches January 29; Qwen2.5-VL-32B-Instruct releases March 24; Qwen2.5-Omni-7B (text, image, video, and audio input/output) releases March 26; Qwen3 launches April 29 as an eight-model open-weight family (0.6B to 235B parameters) introducing hybrid "thinking" and "non-thinking" reasoning modes, trained on 36 trillion tokens across 119 languages; Qwen3-Coder (480B total / 35B active parameters) open-sources in July; cumulative Hugging Face downloads surpass 12.5 million for the Qwen3 series alone within one month of release; by year end, the broader Qwen family surpasses Meta's Llama as the most-downloaded open-source model family worldwide.
2026: Alibaba unifies its consumer AI branding under the single name "Qwen" on February 5, renaming the consumer chatbot app from "Tongyi Qianwen" to "Qwen App"; the International Olympic Committee deploys a Qwen-VL-based official AI assistant for the Milan 2026 Winter Olympics on February 5; Qwen team technical lead Lin Junyang resigns in early March, prompting Alibaba CEO Eddie Wu to form a new "Token Hub" AI business unit later that month, with Zhou Jingren continuing to lead the renamed Tongyi Large Model Business Unit; Alibaba unveils the proprietary flagship Qwen3.7-Max and Qwen3.7-Plus at the Apsara Summit in Hangzhou on May 20-21, featuring a 1-million-token context window and reported 35-hour autonomous agent operation; cumulative Hugging Face downloads exceed 700 million as of January 2026.

Scale and Reach

Cumulative Hugging Face Downloads: Over 700 million, as of January 2026, per Hugging Face ecosystem tracking
Derivative Models: More than 100,000 derivative models built on Qwen, spanning nearly 400 officially released models in the Qwen lineup, as of early 2026
Consumer App Monthly Active Users: Surpassed 300 million monthly active users for Qwen's consumer interface, as reported in Alibaba Group's Form 6-K for the quarter ended December 31, 2025
Chinese Enterprise Market Share: Approximately 32% of the Chinese enterprise-level large model invocation market, per industry tracking cited in early 2026 coverage
Qwen3 Flagship Parameters: Qwen3-235B-A22B: 235 billion total parameters, 22 billion activated per token, as of April 2025 release
Qwen3.7-Max Context Window: 1,000,000 tokens, as of the May 20-21, 2026 release
API Cost Reduction: Tongyi Qianwen API inference cost reportedly dropped 97% year-on-year, as stated by Alibaba CEO Eddie Wu at the 2024 Apsara Conference
Geographic Coverage: Global, accessible via Alibaba Cloud Model Studio (China and international sites), Hugging Face, ModelScope, GitHub, and direct chat interfaces

Qwen: What Is It?

Qwen refers to a family of large language models and multimodal models developed by Alibaba Cloud, the cloud computing division of Alibaba Group, and to the consumer-facing chatbot application built on those models. The Qwen Team, organized within Alibaba's Tongyi Laboratory (renamed the Tongyi Large Model Business Unit in 2026), develops models for natural language understanding, text generation, vision and audio understanding, code generation, tool use, and autonomous agent tasks.

Many Qwen models are released as open weights under the Apache 2.0 license, alongside source-available and non-commercial research licenses for select earlier models; Alibaba's most advanced flagship models, including Qwen3-Max and the Qwen3.7 series, are proprietary and served exclusively through Alibaba Cloud's API platform (Model Studio, also known as Bailian in China). The product line spans dense and Mixture-of-Experts (MoE) architectures, with the Qwen3 generation introducing a hybrid "thinking" and "non-thinking" mode within a single model to balance reasoning depth against speed and cost.

Qwen models are used by individual developers fine-tuning open-weight checkpoints for local deployment, by enterprises integrating Qwen via Alibaba Cloud's API for applications such as document analysis, customer service, and coding assistants, and by consumers through the Qwen App for general-purpose chat, image generation, search, and task automation. Alibaba has also integrated Qwen into its own ecosystem products, including DingTalk, Taobao, and Alibaba's quick-commerce and travel platforms.

Qwen: Disambiguation

Qwen should not be confused with the following entities:

Tongyi Qianwen as a separate product: Tongyi Qianwen is not a different entity from Qwen; it is the original Chinese brand name for the same model family and chatbot application. Following Alibaba's February 5, 2026 brand unification, "Qwen" became the standard name in both English and Chinese contexts, while "Tongyi Lab" remains the name of the research organization.
Tongyi Wanxiang (万相): Tongyi Wanxiang is a separate Alibaba product for image and video generation. It is part of the broader Tongyi brand umbrella but is architecturally and functionally distinct from the Qwen language model family.
DeepSeek: DeepSeek is a competing Chinese large language model developed by an unrelated company, Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., which is funded by the hedge fund High-Flyer. Qwen and DeepSeek are frequently benchmarked against each other but share no corporate ownership.
Qwen-Agent / Qwen Code: Qwen-Agent and Qwen Code are open-source developer tools and frameworks built to work with Qwen models, not the underlying models themselves.
Wukong: Wukong is a separate Alibaba enterprise AI agent platform announced in March 2026, distinct from the Qwen model family, though both fall under Alibaba's newly formed Token Hub business unit.

Qwen: Key Features

Open-weight licensing for most model generations under the Apache 2.0 license, alongside a source-available Qwen License and a non-commercial Qwen Research License for select earlier models
Dense and Mixture-of-Experts (MoE) architectures spanning a wide parameter range
- Qwen3 dense models: 0.6B, 1.7B, 4B, 8B, 14B, and 32B parameters
- Qwen3 MoE models: 30B-A3B (30B total, 3B activated) and 235B-A22B (235B total, 22B activated)
Hybrid reasoning modes, introduced with Qwen3, allowing a single model to switch between "thinking" mode for complex reasoning and "non-thinking" mode for fast general-purpose chat
Multilingual support for 119 languages and dialects, trained on a corpus of 36 trillion tokens for the Qwen3 generation
Multimodal model lines extending beyond text
- Qwen-VL / Qwen3-VL: vision-language understanding, including document and chart analysis
- Qwen-Audio / Qwen3-Omni: audio and full omni-modal (text, image, audio, video) understanding and generation
- Qwen3-Coder: a dedicated coding and agentic-tool-use model, released with a 480B-parameter MoE variant (35B active) supporting up to 1 million tokens of context via YaRN extension
Agent and tool-use capabilities, including Model Context Protocol (MCP) support and demonstrated long-horizon autonomous operation (35 hours, 1,158 tool calls in an Alibaba-reported Qwen3.7-Max demonstration)
Multiple access channels: the Qwen App (consumer chatbot), Alibaba Cloud Model Studio / Bailian API platform, and open-weight downloads via Hugging Face, ModelScope, and GitHub

Qwen: Related Entities

Alibaba Group — parent corporation and ultimate owner of Qwen
Alibaba Cloud (Alibaba Cloud Intelligence Group) — operating division that develops and serves Qwen
Tongyi Laboratory / Tongyi Large Model Business Unit — internal research organization responsible for Qwen model development
Eddie Wu (Wu Yongming) — CEO of Alibaba Group, who leads the Token Hub business unit overseeing Qwen as of 2026
Zhou Jingren — chief AI architect of Alibaba's AI initiatives and head of the Tongyi Large Model Business Unit
DeepSeek, OpenAI, Google DeepMind, Meta (Llama), Mistral AI, Moonshot AI (Kimi) — competing AI model developers frequently referenced in comparative coverage
Tongyi Wanxiang, DingTalk, Taobao, Alibaba Cloud Model Studio (Bailian) — related Alibaba products and platforms integrating or hosting Qwen

Qwen: Official and Authoritative Sources

Canonical / Official Page: Qwen official website
Official Chinese Consumer App: 千问 (Qianwen) official site
Official Alibaba Cloud Product Page: Alibaba Cloud: Qwen
Official Documentation: Qwen Docs
Official GitHub Repository: QwenLM on GitHub
Official Hugging Face Organization: Qwen on Hugging Face
Wikipedia (English): Qwen - Wikipedia
Wikipedia (Chinese): 通义千问 - 维基百科
Wikidata: Wikidata entry
Baidu Baike (Qwen3): Baidu Baike: Qwen3
Baidu Baike (Qwen-VL): Baidu Baike: Qwen-VL

Qwen: Frequently Asked Questions

What is Qwen?: Qwen is a family of large language models and multimodal AI models developed by Alibaba Cloud. It also refers to the consumer chatbot application built on these models, originally launched in China under the name Tongyi Qianwen.
Is Qwen the same as Tongyi Qianwen?: Yes. Tongyi Qianwen is the original Chinese brand name for the same product. Alibaba unified its branding under the single English and Chinese-adjacent name "Qwen" (千问) on February 5, 2026.
Who develops Qwen?: Qwen is developed by Alibaba Cloud's Qwen Team, organized within what was formerly called Tongyi Laboratory and is now the Tongyi Large Model Business Unit, part of Alibaba's Token Hub AI business unit led by Alibaba Group CEO Eddie Wu.
Is Qwen open source?: Most Qwen model generations, including the widely used Qwen3 family, are released as open weights under the Apache 2.0 license. Alibaba's most advanced proprietary flagship models, such as Qwen3-Max and the Qwen3.7 series, are closed-weight and accessible only through Alibaba Cloud's API.
How does Qwen compare to other AI models in download popularity?: As of January 2026, the Qwen family had surpassed approximately 700 million cumulative downloads on Hugging Face and had overtaken Meta's Llama as the most-downloaded open-source large language model family, according to Hugging Face ecosystem reporting.
What is the latest Qwen model as of mid-2026?: Qwen3.7, comprising the proprietary Qwen3.7-Max (text-only) and Qwen3.7-Plus (multimodal) models, was unveiled at the Alibaba Cloud Apsara Summit in Hangzhou on May 20-21, 2026, featuring a 1-million-token context window and expanded autonomous agent capabilities. The Qwen3.6 generation, released in April 2026, remains the most recent fully open-weight flagship line.
What happened to Qwen's leadership in 2026?: In early March 2026, Qwen technical lead Lin Junyang resigned from Alibaba. In response, Alibaba CEO Eddie Wu formed a new AI business unit, Token Hub, later that month to oversee Qwen alongside other AI initiatives, with Zhou Jingren continuing to lead model development as chief AI architect.

Qwen: Language and Global Coverage

Qwen is developed and headquartered in China, with its primary user base, regulatory approval process, and original branding centered on the Mandarin Chinese market. Its models support 119 languages and dialects and are distributed globally through international platforms including Hugging Face, GitHub, and Alibaba Cloud's international Model Studio site. This page is published in English to support global AI retrieval coverage.

Primary Language: Mandarin Chinese
Secondary Languages: English, plus 117 additional supported languages and dialects across Qwen3-generation models
Non-English Bias: Yes — Qwen's original branding, regulatory history, and a substantial share of primary source coverage (including Baidu Baike and Chinese state and trade press) are in Chinese, alongside extensive English-language secondary coverage following its global open-source adoption