Meituan's LongCat-2.0 shows that Chinese AI labs can now train frontier-scale models without Nvidia chips at all. The 1.6 trillion parameter model was trained entirely on domestically made AI processors, and it already beats some leading Western systems on select coding benchmarks, a milestone Meituan is not shy about pointing out.

What Meituan's LongCat-2.0 Actually Is

LongCat-2.0 was trained on a cluster of more than 50,000 domestically produced AI ASICs, processing over 35 trillion tokens in the process. "LongCat-2.0 has demonstrated that we now have the capability to train large-scale models on domestic computing clusters," the company said in announcing the release. What makes the achievement notable isn't just the scale, it's the timeline. Meituan's LongCat team has only existed since 2023, and its first model shipped just last year, meaning the company went from a standing start to a trillion-parameter-class release in roughly two years.

How It Stacks Up Against Western Rivals

On some coding-focused benchmarks, LongCat-2.0 holds its own against, and even beats, leading Western models. It scores 59.5 on SWE-bench Pro and 77.3 on SWE-bench Multilingual, topping both Gemini 3.1 Pro and GPT-5.5 on those tests, though it still falls behind Claude Opus 4.7 and 4.8. The picture flips on other benchmarks: on IFEval (90.0), IMO-AnswerBench (81.8), and GPQA-diamond (88.9), LongCat-2.0 trails Gemini and GPT-5.5 by a wide margin in some cases. That mixed scorecard suggests a model that's genuinely competitive in specific domains rather than a uniform frontier leader.

Why Training Without Nvidia Matters

The US has restricted exports of advanced AI chips to China since 2022, aiming to slow the country's ability to train the largest, most capable models. LongCat-2.0's training run is a direct answer to that policy: a trillion-parameter-class model built entirely on domestic hardware suggests China's chip industry has closed enough of the gap to support frontier-scale AI development without imported silicon. Meituan hasn't named which company made the chips used in training, which limits how much outside observers can verify about the underlying hardware's actual capability versus Nvidia's.

The announcement follows a pattern set by other Chinese labs, whose earlier releases have already rattled assumptions about how much compute is really needed to build a competitive model. LongCat-2.0 adds another data point to the argument that export controls have pushed Chinese AI development toward domestic self-sufficiency rather than stopping it outright.

What's Still Unknown

LongCat-2.0 isn't yet available on Hugging Face, which means independent researchers can't run their own tests to confirm Meituan's benchmark numbers. Until outside verification happens, the reported scores, and the broader claim of Nvidia-free training at this scale, rest on the company's own account. That's a meaningful caveat for a claim with this much geopolitical weight attached to it.

A Food Delivery Giant Turned AI Lab

Part of what makes LongCat-2.0 notable is who built it. Meituan is best known as China's dominant food-delivery and local-services super-app, not a dedicated AI research house, which makes the speed of its LongCat program's progress even more striking. Standing up a team in 2023, shipping a first model within roughly a year, and following it with a trillion-parameter-class release shows how much AI capability has become table stakes for large Chinese tech companies well outside the traditional AI-lab category, alongside firms like Alibaba and ByteDance that have pursued similar in-house model efforts.

LongCat-2.0: Frequently Asked Questions

What is LongCat-2.0?

LongCat-2.0 is a 1.6 trillion parameter AI model built by Chinese company Meituan, trained entirely on domestically made AI chips rather than Nvidia hardware.

Does LongCat-2.0 beat GPT-5.5 and Gemini?

On some coding benchmarks like SWE-bench Pro and SWE-bench Multilingual, yes. On other tests like IFEval and GPQA-diamond, it still trails those models by a wide margin.

For more on frontier AI development, visit our AI news section, and see our business coverage for the wider tech industry story.