Qwen

Qwen is Alibaba Cloud’s generative-artificial-intelligence model and application brand. In Chinese, the broader product identity is commonly written as Tongyi Qianwen, or 通义千问. The brand covers a family of large language models, multimodal models, coding models, and associated assistant and developer offerings rather than a single software release. Its products are developed within Alibaba’s cloud and artificial-intelligence organization and are used both through consumer-facing applications and through application-programming interfaces, cloud services, and downloadable model checkpoints. Alibaba introduced Tongyi Qianwen publicly in 2023 as part of its effort to build a general-purpose AI platform for Chinese and international users. The initial positioning emphasized conversational question answering, content generation, summarization, translation, coding assistance, and enterprise productivity. Qwen models were designed for Chinese and English use, while subsequent releases expanded multilingual coverage, context handling, tool use, coding performance, and multimodal capabilities. The model family has been offered in different sizes so that it can serve both large-scale cloud deployments and more resource-constrained local or edge environments. A significant feature of Qwen’s development strategy has been the publication of open-weight releases alongside proprietary or hosted services. Alibaba has released multiple Qwen checkpoints through public model repositories and developer channels, allowing researchers and companies to run, adapt, and evaluate selected models subject to the applicable license terms. This has helped Qwen become one of the most visible Chinese-origin open-model families and has supported an ecosystem of fine-tuned variants and third-party applications. Open availability does not mean that every Qwen capability or hosted service is unrestricted; licensing, access methods, model size, and commercial-use conditions vary by release. The Qwen portfolio has expanded beyond text-only chat. The Qwen2 generation introduced improvements across model sizes and included specialized variants such as coding-oriented models. Later Qwen2.5 releases broadened the family with stronger general-purpose, mathematical, coding, and multimodal models. The Qwen brand is also associated with Tongyi-based assistant experiences in Alibaba’s consumer and workplace ecosystem, and with model access through Alibaba Cloud’s Model Studio and related platform products. These offerings allow organizations to use Qwen for retrieval-augmented generation, agents, document processing, customer service, software development, and other enterprise workloads. Qwen operates in a highly competitive market that includes Alibaba’s domestic rivals, international foundation-model providers, and open-model projects. Its strategic advantages include Alibaba Cloud infrastructure, access to a large Chinese-language user and enterprise base, integration with Alibaba’s broader software ecosystem, and the ability to distribute both hosted services and downloadable models. Its principal challenges include rapid model obsolescence, computing and energy requirements, safety and reliability expectations, regulatory compliance, intense competition, and the need to maintain consistent performance across languages and deployment environments. The brand remains active and is best understood as a continuing model family and AI platform rather than a conventional standalone company. Public information does not establish a separately incorporated Qwen entity, separately reported financial results, or a standalone stock listing.

Qwen developed from Alibaba’s broader Tongyi generative-AI initiative. Alibaba presented Tongyi Qianwen in 2023 as a general-purpose large language model and positioned it as a foundation for conversational applications, enterprise productivity tools, and integration across Alibaba’s digital ecosystem. The launch occurred during a period in which technology companies worldwide were rapidly introducing generative-AI assistants and foundation models. The early product strategy combined an application layer with a model platform. Consumers could encounter Tongyi-branded assistant experiences, while businesses and developers could access model capabilities through Alibaba Cloud. Intended uses included writing, summarization, translation, question answering, coding, document analysis, and customer-service automation. The emphasis on Chinese-language capability and enterprise deployment reflected Alibaba’s domestic market position, while the use of the Qwen name and distribution through international developer channels supported broader visibility. Qwen subsequently became the name most associated with the underlying model family. Alibaba and the Qwen team released models in multiple sizes, enabling different trade-offs between quality, latency, memory consumption, and deployment cost. Selected model weights and technical materials were distributed publicly, which encouraged independent evaluation, fine-tuning, quantization, and integration into third-party software. At the same time, Alibaba Cloud continued to provide managed access, production infrastructure, customization tools, and enterprise controls for customers that did not want to operate models themselves. The Qwen2 generation represented a major model-family revision in 2024. It covered several scales and was accompanied by improvements in multilingual performance, reasoning-related tasks, coding, and long-context handling. Specialized releases such as Qwen2-Audio and Qwen2-VL extended the family beyond text-only interaction, while Qwen2-Math and related coding variants addressed more focused technical workloads. These releases made Qwen relevant both as a general assistant model and as a component in application-specific systems. Qwen2.5 continued the release cadence with a wider portfolio of general-purpose and specialized models. The family’s positioning increasingly emphasized practical use by developers: models could be hosted through Alibaba Cloud services, downloaded for local experimentation, or adapted through fine-tuning and retrieval-augmented generation. Qwen-Coder and vision-language offerings addressed software development and image-plus-text tasks, while the Tongyi application identity remained connected to end-user assistant products. Qwen’s history reflects Alibaba’s dual approach to generative AI: use proprietary cloud infrastructure and applications to build commercial services, while releasing selected model families openly enough to attract developers and researchers. The brand has therefore operated simultaneously as a consumer-facing assistant identity, an enterprise cloud capability, and an open-model project. Its development has been shaped by competition with Chinese technology companies, international providers, and other open-weight model communities. Public sources do not show Qwen as an independently incorporated company or as a separately listed business; it remains an Alibaba-associated product and technology family.

• 2024

  Qwen2 released

  [3]Qwen Team: Qwen2qwenlm.github.io

  The Qwen team released the Qwen2 family in multiple sizes, with updated language, multilingual, coding, and long-context capabilities.

• 2024

  Qwen2.5 released

  [4]Qwen Team: Qwen2.5qwenlm.github.io

  Qwen2.5 broadened the family with updated general-purpose models and specialized releases for coding, mathematics, and multimodal applications.

• 2023

  Tongyi Qianwen announced

  [1]Alibaba Group: Alibaba Unveils Tongyi Qianwenalibabagroup.com

  Alibaba announced Tongyi Qianwen as a general-purpose large language model for conversational services, enterprise applications, and integration across Alibaba products.

• 2023

  Alibaba Cloud expands model-service access

  [2]Alibaba Cloud: AI Model Service Platformalibabacloud.com

  Alibaba Cloud presented cloud-based model services and tools intended to help organizations use and customize Tongyi-related models.

A Chinese-origin, enterprise-oriented generative-AI platform combining hosted Alibaba Cloud services with a broad family of downloadable and open-weight language, coding, and multimodal models.

• Large language models · 2023

  Qwen

  The core Qwen family consists of transformer-based language models for dialogue, generation, summarization, question answering, translation, reasoning-related tasks, and tool-enabled applications. Models have been published in different sizes and deployment formats, allowing use through hosted APIs or, for applicable releases, local inference and fine-tuning.

• Large language models · 2024

  Qwen2

  Qwen2 is a major generation of the model family released in multiple parameter sizes. It was presented as improving general language performance, multilingual capability, coding, and long-context processing while retaining smaller variants suitable for more economical deployment.

• Large language and specialized models · 2024

  Qwen2.5

  Qwen2.5 is an expanded generation covering general-purpose models and specialist variants. The portfolio includes models oriented toward coding, mathematics, and multimodal understanding, giving developers a choice of models for general assistants and technical workflows.

• Vision-language models · 2023

  Qwen-VL

  Qwen-VL models combine image and text understanding for tasks such as image description, visual question answering, document interpretation, and image-grounded dialogue. They extend the Qwen family from text generation toward multimodal application development.

• Coding models · 2024

  Qwen-Coder

  Qwen-Coder is the coding-oriented branch of the Qwen portfolio. It is intended for code completion, code generation, software explanation, transformation, debugging assistance, and related developer workflows, with distribution through model repositories and cloud-based services.

• AI assistant · 2023

  Tongyi AI assistant

  Tongyi-branded assistant experiences apply Alibaba’s underlying generative models to conversational search, writing, summarization, question answering, and productivity scenarios. Exact features and availability vary by product, market, account type, and Alibaba service.

• Generative AI platform · 2023

  Alibaba Cloud Model Studio

  Model Studio is Alibaba Cloud’s model-development and application platform through which developers and organizations can access foundation models, build generative-AI applications, and use tools for deployment and customization. Qwen is a central model family in the platform’s offering.

• 2024

  Move from Qwen to the Qwen2 generationGeneration change

  Background

  The model family needed stronger capability across languages, coding, long-context tasks, and deployment sizes.

  Action taken

  The Qwen team released Qwen2 models in multiple sizes and added or updated specialized multimodal and technical variants.

  Aftermath

  Qwen broadened from a flagship conversational model into a portfolio serving general, technical, and multimodal use cases.

  Source

• 2024

  Expand the portfolio with Qwen2.5Generation change

  Background

  Rapid progress in foundation models increased demand for stronger general-purpose and domain-specific systems.

  Action taken

  Alibaba’s Qwen team introduced Qwen2.5 and related models focused on general language, coding, mathematics, and multimodal applications.

  Aftermath

  The Qwen brand continued as an actively developed family rather than a single fixed model release.

  Source

• 2023

  Launch Tongyi Qianwen as a general-purpose AI platformProduct launch

  Background

  Alibaba entered the rapidly developing generative-AI market with a model intended to support both end-user assistant experiences and enterprise software integration.

  Action taken

  Alibaba announced Tongyi Qianwen and connected the model initiative to Alibaba Cloud services and the wider Alibaba product ecosystem.

  Aftermath

  The initiative evolved into the Qwen model family, with subsequent public releases, specialist variants, and managed cloud access.

  Source

• 2023

  Combine open model distribution with managed cloud servicesStrategy

  Background

  Developers increasingly expected both downloadable model weights for experimentation and production-grade hosted infrastructure.

  Action taken

  Alibaba made selected Qwen models and related materials available through public developer channels while offering hosted access, customization, and deployment through Alibaba Cloud.

  Aftermath

  Qwen gained visibility among open-model developers while remaining integrated with Alibaba Cloud’s commercial AI platform.

  Source

• 2026.10.31

  Reported2026.11.01

  Alibaba and Apple partner on Qwen integration[7]阿里巴巴与苹果合作整合通义千问tribune.com.pk

  Alibaba partnered with Apple to integrate Alibaba's Qwen AI model into Apple Intelligence for the Chinese market.

• Unconfirmed

  Reported2026.10.13

  Alibaba Qwen releases QwenGyre online reinforcement learning framework[8]阿里千问发布QwenGyre在线强化学习框架163.com

  Alibaba's Qwen team officially released QwenGyre, an online reinforcement learning framework for long-horizon agent tasks, providing elastic scheduling and trajectory handling; in official NL2RepoBench experiments the model pass rate rose from 52.5% to 58.5% after 48 training steps, with a 1.78x end-to-end speedup over the Async implementation under the same GPU budget.

• 2026.08.27

  Reported2026.08.27

  Alibaba launches Qwen3.8-Flash and open-sources Qwen3.8-Flash-Next[9]阿里发布Qwen3.8-Flash并开源Qwen3.8-Flash-Next163.com

  Alibaba released Qwen3.8-Flash, a multimodal MoE model with 125B main parameters and 51B N-gram embedding parameters activating 6B per token, priced at 1 yuan per million input tokens and 3 yuan per million output tokens on the Qwen AI platform, and simultaneously open-sourced the same model as Qwen3.8-Flash-Next on Hugging Face and ModelScope.

• 2026.08.27

  Alibaba Qwen team open-sources Qwen3.8-Flash-Next multimodal MoE model[10]阿里通义千问团队开源Qwen3.8-Flash-Next多模态MoE模型chinaz.com

  Alibaba's Tongyi Qianwen team open-sourced Qwen3.8-Flash-Next, an early preview of the Qwen4 architecture. The multimodal MoE model has 125 billion total parameters with 6 billion activated per token, a 256K native context window extended to 1M via YaRN, and Day-0 support from SGLang.

• 2026.08.20

  Reported2026.08.20

  Alibaba Qwen releases Qwen-UI-Agent GUI agent foundation model[11]阿里千问发布Qwen-UI-Agent图形界面智能体基座模型finance.sina.com.cn

  Alibaba's Qwen large model team officially released Qwen-UI-Agent, a GUI graphical-interface agent foundation model for real devices covering mobile, PC desktop, web and deep-search environments, allowing AI to read screen interfaces and simulate human operation of apps and software. The model reported benchmark scores including 82.1% on MobileWorld, 92.2% on MobileWorld-Real, 97.5% on AndroidDaily, 79.5% on OSWorld-Verified, 73.6% on WebArena and 81.5% on ScreenSpot-Pro, and includes safety mechanisms requiring user confirmation for sensitive operations such as payments, data deletion and permission grants.

• Unconfirmed

  Reported2026.08.15

  Alibaba open-sources Qwen3.8-27B model[12]阿里开源 Qwen3.8-27B 模型163.com

  Alibaba open-sourced Qwen3.8-27B, a native multimodal dense model with 27 billion parameters that it says surpasses Qwen3.7-Plus overall and performs well in real coding and office scenarios, running smoothly on consumer-grade GPUs after quantization.

• 2026.08.15

  Reported2026.08.18

  MediaTek and Qwen deepen cooperation with Day-0 support for Qwen 3.8 on Dimensity platforms[13]联发科与千问深化合作，天玑平台实现 Qwen 3.8 Day-0 支持163.com

  On August 15, MediaTek announced it is deepening cooperation with Qwen, with its Dimensity automotive cockpit platform C-X1 and Dimensity flagship chips achieving Day-0 support for Qwen 3.8, allowing devices on Dimensity platforms to access the new Qwen model faster.

Read more

1. 1Alibaba Group: Alibaba Unveils Tongyi Qianwenalibabagroup.com
2. 2Alibaba Cloud: AI Model Service Platformalibabacloud.com
3. 3Qwen Team: Qwen2qwenlm.github.io
4. 4Qwen Team: Qwen2.5qwenlm.github.io
5. 5Qwen Official Project Siteqwenlm.github.io
6. 6Alibaba Cloud Model Studioalibabacloud.com
7. 7阿里巴巴与苹果合作整合通义千问tribune.com.pk
8. 8阿里千问发布QwenGyre在线强化学习框架163.com
9. 9阿里发布Qwen3.8-Flash并开源Qwen3.8-Flash-Next163.com
10. 10阿里通义千问团队开源Qwen3.8-Flash-Next多模态MoE模型chinaz.com
11. 11阿里千问发布Qwen-UI-Agent图形界面智能体基座模型finance.sina.com.cn
12. 12阿里开源 Qwen3.8-27B 模型163.com
13. 13联发科与千问深化合作，天玑平台实现 Qwen 3.8 Day-0 支持163.com
14. 14阿里开源发布Qwen3.8-Flash，训练成本骤降近90%163.com
15. 15国行iPhone重磅更新！iOS 27.3 Beta 1发布：终于等来苹果AI 支持实时翻译等finance.sina.com.cn
16. 16Qwen 3.8发布即适配，联发科端侧AI再展“Day-0速度”finance.sina.com.cn