Editorial illustration for Alibaba's Qwen2.5 Debuts Over 100 Open-Source AI Models
Alibaba Qwen2.5 Launches 100+ Open-Source AI Models
Alibaba Cloud started small, handing out invitation codes. On April 7, 2023, the company began letting corporate customers test a chatbot called Tongyi Qianwen, a name that loosely translates to "truth from a thousand questions" and nods to Mencius. Four days later, then-CEO Daniel Zhang showed it off at the Alibaba Cloud Summit in Beijing, promising to wire it into DingTalk and the Tmall Genie assistant.
That was the cautious version. The real shift came four months later, when Alibaba open-sourced Qwen-7B on August 3, 2023, trained on more than 2.2 trillion tokens, built to compete directly with Meta's Llama 2. A vision-language branch, Qwen-VL, arrived weeks after.
By September, the chatbot was public and the team had a technical report on arXiv. By December, Alibaba had pushed out both a 72B and a 1.8B model, covering everything from laptop-friendly to frontier-scale.
That was just the first year. What Alibaba built next, release after release, is what turned Qwen from a regional chatbot experiment into a model family with open weights scaling into the trillions of parameters.
In April 2023, Alibaba Cloud demoed a chatbot whose name roughly means ‘truth from a thousand questions.’ Three and a half years later, its descendant ships open weights with 2.4 trillion parameters. This is the story of how Qwen got there, release by release.
Why this matters
For anyone building on open models, the Qwen timeline is the actual story of this AI cycle: not one lab's breakthrough, but a steady compounding of pretraining data (7 trillion to 18 trillion tokens) and parameter counts (7B to 2.4T) that turned a corporate chatbot demo into one of the broadest open-weight catalogs available. Dropping over 100 models in a single day at Apsara is a scale of release cadence most Western labs don't attempt, and it matters for developers because it lowers the cost of picking the right-sized model for a job rather than defaulting to whatever a closed API offers. Researchers get a rare paper trail, the technical report, to study how data scaling alone moved capability.
Founders should note the trajectory more than any single release: Alibaba shipping video-length multimodal models (Qwen2-VL) and trillion-parameter open weights within the same year signals Chinese labs competing on raw output volume, not just benchmark wins. Worth watching next: whether that pace holds as parameter counts climb, or whether Alibaba starts trading breadth for fewer, more deliberate releases.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
From Archives OpenAI’s Sol Matches Astra Performance at a Fraction of the Cost
From Archives OpenAI Adds Subtle Text Watermark to ChatGPT in European Union
Common Questions Answered
What does the name Tongyi Qianwen mean and why did Alibaba choose it for their chatbot?
Tongyi Qianwen loosely translates to 'truth from a thousand questions' and references the ancient Chinese philosopher Mencius. Alibaba Cloud chose this name when they first introduced the chatbot to corporate customers in April 2023, reflecting the philosophical approach to seeking truth through inquiry.
How much has Qwen's model capacity grown from its initial release to Qwen2.5?
Qwen has grown dramatically from 7 billion parameters in its early versions to 2.4 trillion parameters in Qwen2.5, representing a massive expansion in model capability. Additionally, the pretraining data increased from 7 trillion tokens to 18 trillion tokens across the development timeline.
Why is Alibaba's release of over 100 open-source models significant for the AI development community?
Alibaba's release of over 100 open-weight models in a single day at Apsara represents an unprecedented scale of release cadence that most Western AI labs do not attempt. This broad catalog of open models provides developers with diverse options for building applications and demonstrates a commitment to open-source AI development.
What was the initial strategy Alibaba Cloud used to introduce Tongyi Qianwen to the market?
Alibaba Cloud started cautiously by distributing invitation codes to corporate customers beginning April 7, 2023, allowing them to test the chatbot. The company then demonstrated it publicly at the Alibaba Cloud Summit in Beijing four days later, where then-CEO Daniel Zhang promised to integrate it into DingTalk and the Tmall Genie assistant.
What does the progression of Qwen models reveal about the AI development cycle according to the article?
The Qwen timeline demonstrates that the AI cycle is driven by steady compounding improvements in pretraining data and parameter counts rather than single breakthrough moments from one lab. This consistent evolution from 7B to 2.4T parameters transformed Alibaba's initial corporate chatbot demo into one of the broadest open-weight model catalogs available.
Further Reading
- Alibaba's Qwen Powers Top 10 Open-Source Models as China AI Know-How Goes Beyond DeepSeek - South China Morning Post via Yahoo News
- Alibaba Diverges From Baidu and OpenAI's Closed-Source Strategy, Releases 100+ Open-Source AI Models - CCN
- Alibaba Releases 100+ Open-Source AI Models and New Text-to-Video Generator - SiliconANGLE
- Qwen2.5 Technical Report - arXiv
- Alibaba Cloud Unveils Qwen2.5, Full-Stack AI Infrastructure and Open-Source Models - Alibaba Cloud
