Ant Group is the Chinese financial-technology company behind Alipay, one of the world’s largest mobile payment platforms, and — since 2025 — a fast-growing family of AI models called Ling. It was spun out of Alibaba in 2011, remains roughly a third owned by it, and has spent the past two years building its own large language models so that its payments, lending, and merchant-services businesses can run AI agents instead of app menus and call-center staff.
From Alipay to Ant Group
Alipay started in 2004 as an escrow tool Alibaba built so buyers and sellers on its Taobao marketplace could trust each other with money before a package arrived. Jack Ma spun the payments business out into its own company in 2011, and it grew first into Ant Financial and then, in 2020, into Ant Group — a broader group that now includes Alipay itself, the consumer-lending product Huabei, the online bank MYbank, and the credit-scoring service Zhima Credit.
Ant Group tried to go public in a $34 billion dual listing in Shanghai and Hong Kong in November 2020. It would have been the largest IPO in history, but Chinese regulators halted it days before trading was due to start, shortly after Ma publicly criticized the country’s banking rules. The company spent the following years restructuring under regulatory pressure: it was fined nearly $1 billion for governance and money-laundering violations in 2023, and in 2024 its controlling shareholders, including Ma, gave up the voting rights that had let him steer the company directly. Alibaba still owns about a third of Ant Group today, and the two remain commercial partners, but they run as separate companies with separate strategies — including separate AI models.
Why a payments company builds its own AI
Running Alipay, a lender, and an online bank means handling enormous volumes of customer support, fraud checks, and merchant onboarding — work increasingly done by AI agents rather than fixed menus or human staff. Rather than license models from providers largely walled off in China, or rely entirely on sibling company Alibaba’s own Qwen models, Ant Group builds its own line, called Ling (灵), through a research group it operates named InclusionAI.
Ant’s models are tuned specifically for agent work — multi-step tasks like reconciling a merchant’s transactions, walking a customer through a loan application, or writing and testing code — rather than open-ended chat. Its newest release, Ling-3.0-Flash, is built as a fast “execution” layer that larger, slower reasoning models can hand tasks off to, a division of labor Ant calls a planning-execution separation approach.
Cheap by design, not by accident
What sets Ling apart technically is how few of its parameters switch on for any given request. Ling-3.0-Flash has 124 billion total parameters but activates only about 5.1 billion of them per token, using a mixture-of-experts design that routes each query to a small subset of specialized sub-networks instead of the whole model. That keeps responses fast and computing costs low while letting the model match systems two to three times its active size on reasoning and coding benchmarks — the same efficiency-over-raw-scale bet that other Chinese labs, including DeepSeek and Moonshot, have made with their own models.
Ant also publishes its Ling models as open-weight files on Hugging Face, through its inclusionAI page, for any developer to download, run, and modify. That trades direct licensing revenue for wider adoption and influence over which models other companies build their own products on.
Why it matters
Ant Group’s AI push matters mostly because of scale: Alipay has over a billion users, and any agent Ant deploys inside it touches a huge slice of everyday commerce and personal finance in China. It’s also a signal about where AI is heading — a company built on payments infrastructure, not model research, has decided it needs to own its AI stack rather than rent someone else’s, a pattern now repeating at banks, retailers, and telecoms well beyond China.
In the news
Ant Group’s newest model, Ling-3.0-Flash, made headlines this week for matching the performance of much larger systems at a fraction of the computing cost — read more in Ant Group Unveils Ling-3.0-Flash, an Efficient AI Model for Agents.