Is Alibaba's Qwen the Open-Source AI Winner?
A deep dive into Alibaba’s AI strategy, the company behind the Qwen series
There are so many things to be excited about in the near future of AI as we head into 2025 in just a few short weeks. Things I’m especially watching are Agentic AI, text-to-video and open-source model capabilities. I expect some surprises from China in these domains next year. This piece was first published on November 13th, 2024. The entire piece goes live to all readers on December 6th.
About that Open-source trend, recently Google DeepMind open-sourced AlphaFold 3. Even more impressive to me was the recent release of Qwen 2.5 Coder models, the world doesn’t just need more benchmarks after all. Epoch AI went to the trouble of collaborating with 60+ leading mathematicians to create hundreds of original, exceptionally challenging math problems, of which current AI systems solve less than 2%. Alibaba Cloud’s Qwen was already something I wanted to cover for my readers, as it’s under-appreciated according to many of the top Open-source experts I watch. The Qwen models and also like Meta’s Llama models are showing the huge potential of open-source LLM innovation.
“Qwen2.5 is so far ahead of other LLMs that it has become normal not to include it in evaluations.” - Benjamin Marie (October 28th, 2024).
I reached out to the super talented
(Grace Shao, ex CNBC) to dive into Alibaba, who I consider an expert on to dig into its somewhat mysterious prowess in open-source LLM building. That’s how this guest post came into being. Check out her Newsletter AI Proem, a four month old publication that I recommend and came on my radar:AI Proem
AI at the intersection of geopolitics, energy, society with special corporate deep dives.
In this short video Grace introduces her Newsletter AI Proem and the evolution of her work.
AI in Geopolitics 🌍
AI in Energy ☀️
AI in Society 🏙️
I really like the topics she writes about, and if you are at all like me, you probably will too.
A lot of my own work with guest writers is uncovering unique talents and emerging writers like Grace. Her experience at Alibaba and with CNBC and now as a writer and researcher suggests a unique vantage point to help us better understand Qwen and Alibaba’s renewed commitment to AI. This sort of coverage is moreover unavailable to most Western media outlets and traditional publications who for whatever reason might want to downplay how innovative China’s AI community actually is. I think that’s a mistake. No matter where innovation is occuring, we should be covering it.
“Qwen 2.5 Coder Just Blew SOTA Closed LLMs Out Of The Water. The folks on the other side of the world in China are not just geniuses; they are big proponents of open-source as well - Tweet by Bindu Reddy [source].”
I was already really primed to think highly of the Qwen Models in the Generative AI open-weight spectrum. Following people like Omar Sanseviero of Hugging Face and others on his team,
(Max Labonne, who now works at Liquid AI), (Elvis Saravia, who has a decent YouTube channel as well), (Benjamin Marie) and many others, I knew the Qwen Models were more important than the lack of Western tech news I was getting on it. But it seems my timing was good, since their last release really ups the bar. So about that recent release:The Latest Qwen Model Surpasses some Proprietary Closed-Source Models
On November 11th, 2024 Alibaba Cloud’s Qwen team released Qwen2.5-Coder-32B-Instruct. The performance of this model surprised a lot of people but not those of us who were already bullish on Qwen. The flagship model, Qwen2.5-Coder-32B-Instruct, reaches top-tier performance, highly competitive (or even surpassing) proprietary models like GPT-4o, in a series of benchmark evaluation, including HumanEval, MBPP, LiveCodeBench, BigCodeBench, McEval, Aider, etc. It reaches 92.7 in HumanEval, 90.2 in MBPP, 31.4 in LiveCodeBench, 73.7 in Aider, 85.1 in Spider, and 68.9 in CodeArena.
Open-source (technically Open-weight) large language models are an exciting domain because they can lead to a lot of real-world applications and a democratization of Generative AI globally. The idea and continuous debate of how many months behind Open-source models are to proprietary frontier and closed-source models really fascinates me and the axis between centralization and decentralization as well. China BigTech needs to follow the enormous capex investments of the West if they can. Thus Tencent and Alibaba are upping their game with Baidu now and Alibaba Cloud’s Qwen2.5-Coder-32B-Instruct feels to me almost like a new paradigm for the ceiling of Open-Source LLMs heading into 2025.
If you are interested in keeping close tabs on Open-source LLM developments, the single best person to follow is Omar from HuggingFace on Twitter/X here.
The guest author today is actually rather distinguished. Grace writes commentaries for Fortune, The Diplomat and other international publications on AI, tech and corporate governance. Grace is an alumna of Columbia University, Tsinghua University, and the Alberta School of Business.
Top 5 reads from Grace:
Let’s review some of her best recent pieces: (these are the titles)
AI arms race far from over: chips is only half the game, infrastructure is the other
Johor, Malaysia: AI Data Centers Drive Investment, Infrastructure Challenges Persists
Read the Technical Report (32 PDF pages)
Related Links to Qwen2.5 Coder-32B-Instruct
Dive even deeper into Qwen.5 Models, read the supporting documents:
Tech Report: https://arxiv.org/abs/2409.12186
The new Qwen2.5 in Code capabilities reach SOTA for open-source models. It’s a threshold that shows China BigTech’s commitment to Generative AI and model building is real and accelerating quickly now in late 2024. In other words, it matches GPT-4 coding capabilities, but it’s Open source & with an Apache 2.0 license, it supports 40+ programming languages and comes available in 6 distinct sizes (0.5B-32B).
As an international media consultant, Grace has also advised clients including Lenovo, Ant Financial, Kuaishou, PayPal and growth-stage consumer tech and biotech firms. Grace was an integral member of Alibaba Group's international corporate affairs team based in Hong Kong, helping the tech giant navigate through some of its largest crises and milestones.
I hope you enjoy the deep dive! I will go into more details about Qwen’s new models and what they mean on one of my related Newsletters soon. I am really look forward to Grace’s coverage as an independent researcher and writer unveiling China BigTech and deeper insights than have previously been easily made available to western audiences. This piece should help you understand China BigTech better in a nutshell as well, so it’s an excellent primer if that interests you. Following voices like
(Tony Peng), (Tech Buzz China Insider), there weren’t previously many insider voices about China Tech on this platform. (Baiguan China Insights) and many writers bring context to China’s role in the world, but few dive into its AI and Tech stack well. Getting the technology category here to offer more balanced coverage, is something I really support.By
(Grace Shao, October, 2024).A deep dive into Alibaba’s AI strategy, the company behind the Qwen series
Despite chip bans from the U.S., China’s AI ecosystem has outperformed many expectations, especially garnering international attention from developers as Alibaba’s open-source Qwen series have been widely adapted and been widely talked about in the community.
China has created a completely separate AI ecosystem for various reasons: 1) to lessen dependency on the West 2) the Great Firewall censorship constraints; but that is not to say that innovation is stifled. In fact, there is an extremely vibrant set of players in China right now across the AI ecosystem, and today we will dive deep into the role Alibaba plays in the space.
Screenshot of graph prepared by Grace on China’s major BigTech players.
Alibaba
Alibaba Cloud offers robust cloud infrastructure with support for open-source models and extensive AI services.
Qwen-72B and Qwen-1.8B are advanced LLMs developed by Alibaba Cloud, with capabilities in multimodal processing.
Dingtalk, enterprise chat platform
Alimama, AI-driven ad-optimization tool set for SMEs selling on the Tmall and Taobao.
Tencent
Tencent enhances its AI capabilities through its Intelligent High-Performance Network, optimizing GPU usage for LLM training.
Hunyuan is Tencent's in-house LLM aimed at enterprise applications, with a focus on efficiency and cost-effectiveness.
Tencent's AI services include personalized news feeds and chatbot solutions across its existing apps.
Huawei
Huawei Cloud provides a high-performance infrastructure tailored for AI applications, focusing on self-reliance in technology.
Pangu 3.0 consists of foundational, industry-specific, and scenario-specific models designed for diverse applications across sectors.
Huawei's LLMs are used in various industries such as finance and healthcare to enhance digital transformation efforts.
ByteDance
ByteDance leverages its cloud infrastructure to support the deployment of its LLMs, emphasizing cost efficiency in AI services.
Doubao is a family of LLMs launched by ByteDance, designed for various applications with aggressive pricing strategies.
Applications like Doubao Chatbot and other generative AI tools are aimed at enhancing user interaction and content generation.
Baidu
Baidu Cloud provides comprehensive infrastructure for AI model training and deployment, focusing on technological advancements in AI.
Ernie is Baidu's flagship LLM that has seen significant improvements in training efficiency and application performance over time.
Baidu's applications utilize Ernie for enhanced search capabilities, conversational agents, and other AI-driven solutions.
Leadership Reshuffle: Alibaba Doubles Down on AI as Joe Tsai Returns
Most of you know about Alibaba as the ecommerce company that was founded by the flamboyant billionaire Jack Ma, but Alibaba actually has a wide range of businesses. The ecommerce juggernaut was once the unquestionable market leader, but it faced some fierce competition from newcomers such as Pinduoduo and ByteDance’s Douyin over the years.
This time around, it is making sure that it will not be complacent, and is going all in on AI with cloud computing, its own proprietary LLM model, innovation in AI application and investment into the whole ecosystem, as “AI is too important of a path to just go one direction,” said Joe Tsai, Chairman of Alibaba Group at a recent JP Morgan event.
For a while, there were doubts about the company’s future after the founder’s notorious Shanghai Bund talk that led to a series of high profile government probes into the business. Two years of volatility ended with a series of internal personnel changes amid a slowing domestic consumption environment. And the company’s share price tumbled to USD 57 in October 2022, the lowest it’s ever been.
For context, the previous CEO Daniel Zhang rose from the ranks of Alibaba’s corporate ladder and mostly prioritized the e-commerce business. He was notably known for creating Alibaba’s Singles Day and as the architect of its international e-commerce strategies and innovations. He was always very pragmatic and focused on what he often referred to as the "flywheel models”, to describe Alibaba's growth strategy and operational dynamics. What he meant in layman terms was a self-reinforcing cycle where the interconnectedness of its e-commerce platforms, data-driven insights, and continuous reinvestment in technology and services would create synergy and thus sustainable growth and competitive advantage against others in the race. Although cloud was a key business for him, it was mostly seen as a computing service provider, and it wasn't until more recently, did Alibaba more publicly come out to be an advocate for AI integration and putting cloud front and center of its business strategy.
Video of Joe Tsai speaking at the JP Morgan Global China Summit .
After nearly ten years at the realm, in late 2023, Daniel Zhang stepped down from his CEO position and passed over the baton to Eddie Wu. And Joe Tsai reemerged in the public. This was a huge turning point for the company’s AI strategy.
Joe Tsai’s sudden return to the front and center of the company’s daily operations in 2023 came as a surprise to many. When he returned as the Chairperson of Alibaba Group, he seemed to have brought back his business intuition as well as his observations from California – where he was residing for most of COVID – back to the Hangzhou-based company.
Joe has always been a compelling spokesperson, a charismatic speaker and storyteller, but when he was partnered with Jack, he was seen as the “posh and relatively more reserved one” where Jack was the company’s visionary, spiritual leader and loudest advocate. However, now things have changed, in contrast to his new partner in crime Eddie Wu, who is reserved and media-shy, Joe is happily telling the company’s story, going on podcasts and panels across the region to boost investor confidence, especially highlighting the company’s ambitious AI strategy.
Joe has so much confidence and conviction in the company’s new found focus that he has apparently now even started a mantra internally, hyping employees up with a chant: “Baba will reach 200 again,” which implies that $BABA’s stock price will shoot up to $200 again, (in the last two years it dropped from its peak of over USD300 to ~USD100).
Alibaba’s AI Playbook
Image: Alibaba is invested in AI in five major ways with a twin-strategy,
End-to-End tech stack strategy
Building proprietary LLM - Qwen and offering its LLMs to AI builders
Cloud computing service
Designing chips catered for processing AI applications.
Ecosystem Strategy
Implementing AI into its existing consumer facing applications
Funding AI companies across the ecosystem
Alibaba is easily the most well-known Chinese tech company internationally with a leading cloud business and its own proprietary LLM technology. Although in China, Baidu and Huawei each have their own models and cloud service, Baidu’s data focused strategy has always been more focused on its autonomous driving technology and Huawei has always been more focused on compute and hardware, whereas its LLM is seen to be more a “nice to have” add on for enterprise clients.
In contrast, Alibaba has repeatedly said that it aims to “make AI accessible to all.” At the 2024 Apsara Conference, Alibaba CEO Eddie Wu emphasized that the company is committed to supporting the open-source ecosystem from chips, servers, networks, to storage and data centers.
Proprietary LLM: Tongyi Qianwen (Qwen)
At the forefront of Alibaba's AI offerings is Tongyi Qianwen, a large language model akin to a "super chatbot." This advanced model is capable of understanding and generating text, making it suitable for a wide range of applications, including article generation, conversational responses, and customer support.
The Qwen series - have incredible scale, performance across benchmarks, multimodal features, and commitment to accessibility for a wide range of users. And Alibaba has made this technology publicly available, allowing other businesses to utilize it for free to enhance their customer service capabilities.
“It is the most-competitive Chinese LLM when compared to the likes of GPT4/4.o in terms of its overall performance,” said Leo Jiang, founder of GroundAI and former Huawei Chief Digital Officer.
He added, what makes Qwen special is because of its two formats, “its API driven LLM service offers quicker time to market, and cost effectiveness. Whereas its open-source version gives more control and privacy to its clients.”
Alibaba launched its large language development tool Tongyi Qianwen in 2023 and it is often referred to as Qwen and it is now at its 2.5 iteration. The Qwen models, including the Qwen-72B and Qwen-1.8B, are notable for their diverse parameter sizes—ranging from 1.8 billion to 72 billion parameters—and their multimodal capabilities, which allow them to process not just text but also audio and visual data. This flexibility is enhanced by their training on over 3 trillion tokens, enabling them to outperform many other open-source models across various benchmarks, including multitask accuracy and code generation capabilities.
Editor’s Note:
“An open LLM that competes with Anthropic (AI) Claude Sonnet 3.5 impossible? No, Alibaba Qwen’s 2.5-Coder-32B entered the game matching Claude Sonnet 3.5 across multiple coding benchmarks. Early testers say, “doing the things Sonnet didn't want to,” “Trying out a preview of Qwen2.5 Coder 32B, and it feels like Claude 3.5 Sonnet”., “I am blown away how well it does in long context work.” - Philipp Schmid, Hugging Face
Benchmarks showing Qwen2.5 Coder 32B Instruct performs vs. GPT-4o and Claude 3.5 Sonnet:
Qwen has positioned itself as an all-around AI assistant, with five key application use-cases:
1) real-time meeting transcription and summaries
2) processing lengthy content and providing summaries that require complicated comprehension
3) AI PowerPoint presentation creation
4) real-time simultaneous translation
5) video chat with an AI agent that can provide problem solving.
Image Source: Alibaba
The uniqueness of Qwen lies in its impressive technology and strong commitment to open-source principles, as Alibaba makes various versions of its models available on platforms like Hugging Face and ModelScope. This approach fosters a collaborative environment where developers can experiment and innovate, democratizing access to advanced AI technologies for businesses of all sizes.
In particular, companies with fewer than 100 million monthly active users can use these models for free, promoting wider adoption across industries. And by supporting the growth of the open-source community, Alibaba has aimed to empower users to effectively harness AI capabilities while reducing reliance on proprietary technologies.
ChinaAI’s Jeff Ding translated the well-circulated AItechtalk article on why Qwen is the world’s most popular open-source large model right now, which wrote that, “per Hugging Face data, the Qwen series/bloodline of models has reached more than 50,000. That is, developers around the world have trained more than 50,000 derivative models based on the Qwen series base, second only to the Llama series of about 70,000. This data is the most convincing indicator for judging the ecosystem-level influence of a model.” (
of ChinAI Newsletter).Impressively, the Qwen models have garnered significant interest from across sectors, including automotive, gaming, and scientific research last year. The models have been downloaded over 40 million times since their introduction. Additionally, the lightweight Qwen-1.8B model is designed for deployment on edge devices such as smartphones, making it an attractive option for applications requiring lower computational resources.
The most recent comprehensive upgrade of Qwen2.5 means larger parameter scale, more powerful comprehension of photos and videos, a large-scale audio language model and continued open source models. Not only has it been improved drastically, the cost of strong inference capabilities to support complex tasks have been reduced for both Qwen-Plus and Qwen-Turbo.
Looking ahead, CEO Eddie Wu noted that while AI development has progressed rapidly, AGI (Artificial General Intelligence) is still in its early stages. He emphasized the importance of collaboration and highlighted that the API inference cost for Tongyi Qianwen has dropped by 97% year-on-year, a key factor contributing to its growing popularity. In fact, this is verified by Leo, the former Huawei executive who noted that the Qwen models offer higher accuracy and factuality compared to most other models based in China. It can be customized for enterprise use cases that prioritize the accuracy of outputs and aim to minimize model hallucinations and in addition, Qwen’s biggest edge right now is that it is providing developers a powerful yet cost-effective alternative.
Alibaba Cloud
AI and the cloud business is like the left hand and the right hand, said Joe Tsai in a podcast speaking with Norwegian hedge fund manager Nicolai Tangen. As mentioned earlier, anyone can use Alibaba’s LLM through APIs, or directly go to its open-source model. However, for any of them who want to deploy Qwen they would need cloud computing power and Alibaba Cloud is there to provide that.
In fact, currently 80% of China’s technology companies and half of the country’s large model companies run on Alibaba Cloud. This scale is simply unmatchable. Joe reiterated that with its cloud service as the largest provider in APAC, Alibaba has a huge advantage in garnering data and trial for its Tongyi Qianwen. The positive cycle allows the two businesses across the AI layers to continuously feed into each other.
In addition, the company has created the largest open-source community called ModelScope which hosts many other open-source models on the marketplace and when developers use those open-source models, they will also need compute power, which has become a main driver for Alibaba’s cloud revenue.
By providing the cloud infrastructure to the startups, the tech giant is hoping to hedge its bets by allowing them to access the best consumer-facing application first hand. Providing the cloud infrastructure would enable the company to access a diverse pool of data across domains and use cases which it could potentially leverage to finetune its own models if given the permission. It would also mean talent acquisition and exposure to new innovation in the field will be more accessible.
Alibaba’s AI Applications
So let’s take a look at the application front. Alibaba has integrated AI into its own operations extensively, utilizing it for product recommendations on its e-commerce platform, intelligent customer service, AI empowered advertisement targeting, and AI-driven solutions in cloud services. Additionally, it is looking for ways to better use AI to enhance logistics efficiency and other use cases as well. Today, let’s just take a look at a few mature ones first.
The Artificial Intelligence Online Serving (AI OS) is a platform developed by the company's search engineering team. AI OS integrates personalized search, recommendation, and advertising, supporting various business scenarios across Alibaba's platforms, mostly focusing on the marketplace apps such as Taobao. The technology originally focused on Taobao's search capabilities has expanded to include deep learning technologies and various engines for search and recommendation.
Dingtalk is an enterprise chat software, similar to Slack. Across Dingtalk, all products have been AI-enabled with an embedded AI agent for enterprise and personal use which was launched at the beginning of 2024. The AI agent is a virtual robot that can examine data analytics and is equipped with memory, planning and execution capabilities.
The format to interact with the agent is through a chatbot similar to ChatGPT. The company’s suggested use cases include using the robot as a sales person, IT, HR administrative, financial or procurement staff and it can help companies automate many of the repetitive tedious tasks within the management process.
Meanwhile, Alimama is a platform that helps brands with ad-optimization on Alibaba’s ecommerce marketplace apps - Tmall/ Taobao. Alimama is a relatively unknown business unit of Alibaba but it was actually founded very early on in 2007. It is a digital marketing platform for businesses that are selling on the Taobao or Tmall platforms. The AI-empowered multi-media LMA was launched in April this year and has been fully applied to 2B applications now. The tools include AI sales agents that can handle client enquiries and basic ad design functions that can help improve efficiency and quality, sales analytics for budgeting and pricing purposes and inventory management which have all contributed to an improvement in ROI and text to picture or video generation ad services is also provided by Alimama with relatively low costs. And the company claims to have served over 1 million merchants on the platform and significantly reduced advertising production costs.
Investing to Capture All Possibilities (Opportunities)
[The rest of this article is available to premium readers only and includes explorations into].
Alibaba’s investments in Generative AI
Alibaba’s AI chips: T-Head
Keep reading with a 7-day free trial
Subscribe to AI Supremacy to keep reading this post and get 7 days of free access to the full post archives.