# Meituan open-sources LongCat-2.0: 1.6T model, trained without Nvidia

> Meituan open-sourced LongCat-2.0 in late June 2026 — a 1.6T MoE trained end-to-end on Chinese chips, no Nvidia.

*The brief: a food-delivery giant gives away a trillion-parameter AI built entirely on Chinese chips — and the benchmarks are its own.*

By The FeaturedDaily Desk · FeaturedDaily
Canonical: https://featureddaily.com/news/meituan-longcat-2-brief

> **Key:** **The brief:** Meituan gave away a 1.6-trillion-parameter model under MIT and says it trained the thing without a single Nvidia chip. That supply-chain claim, not the razor-thin benchmark edge over GPT-5.5, is the durable number here.

## What happened

In late June 2026 **Meituan** — China's largest food-delivery and local-services platform — open-sourced **LongCat-2.0** under a permissive **MIT licence**. It's a **1.6-trillion-parameter** Mixture-of-Experts model with a native **1-million-token context**, activating roughly 33-56 billion parameters per token. Meituan also revealed it was the model behind **'Owl Alpha'**, the anonymous stealth entry that had topped OpenRouter's developer usage charts for about two months. The headline claim: it's the **first trillion-parameter model trained and served end-to-end on Chinese-made chips** — a cluster of more than 50,000 domestic ASICs, no Nvidia GPUs.

> Meituan open-sourced LongCat-2.0 as a 1.6-trillion-parameter model trained entirely on Chinese chips, and confirmed it was the same system that had been leading OpenRouter's developer charts anonymously as 'Owl Alpha'.
> — [VentureBeat](https://venturebeat.com/technology/meituan-open-sources-longcat-2-0-the-1-6t-near-frontier-agentic-coding-model-thats-been-leading-openrouter-trained-entirely-on-chinese-chips), 2026-06-30

## The catch

Read the wins carefully. The benchmarks are **Meituan-reported and not independently verified**: 59.5 on SWE-Bench Pro (vs GPT-5.5's 58.6), 70.8 on Terminal-Bench 2.1, 77.3 on SWE-Bench Multilingual. The much-repeated 'beats GPT-5.5' line holds on **exactly one test by under a point** — a statistical tie, not a settled ranking. At announcement Meituan also listed the **full weights as 'coming soon'** rather than all posted, and the non-Nvidia training claim is the company's own until someone reproduces it.

| Benchmark | LongCat-2.0 (reported) | GPT-5.5 |
| --- | --- | --- |
| SWE-Bench Pro | 59.5 | 58.6 |
| Terminal-Bench 2.1 | 70.8 | — |
| SWE-Bench Multilingual | 77.3 | — |

## What's next

Watch three things: whether the full weights ship as promised, whether independent evaluators confirm the coding scores, and whether anyone outside Meituan can reproduce the no-Nvidia training claim. The wider signal is already clear — Chinese open-weight models now make up roughly **61% of OpenRouter's top-10 traffic**, and LongCat-2.0 is the heaviest entry yet in that surge.

## Key takeaways

- Meituan open-sourced LongCat-2.0 in late June 2026 — a 1.6-trillion-parameter Mixture-of-Experts model under an MIT licence, with a native 1M-token context.
- Meituan says it's the first trillion-parameter model trained and served end-to-end on Chinese-made chips: 50,000-plus domestic ASICs, no Nvidia GPUs.
- It was previously the anonymous 'Owl Alpha' that topped OpenRouter's developer charts for about two months before the reveal.
- Benchmarks are Meituan's own and unverified: 59.5 SWE-Bench Pro (vs GPT-5.5's 58.6) — a sub-one-point edge on a single test.

## FAQ

### What is LongCat-2.0?
A 1.6-trillion-parameter open-weight Mixture-of-Experts AI model that Meituan open-sourced in late June 2026 under an MIT licence, with a native 1M-token context. It activates roughly 33-56 billion parameters per token.

### Was it really trained without Nvidia?
Meituan says so — it describes LongCat-2.0 as the first trillion-parameter model trained and served end-to-end on Chinese-made chips, a cluster of more than 50,000 domestic ASICs with no Nvidia GPUs. The claim is Meituan's and had not been independently verified at launch.

### Does it beat GPT-5.5?
Only on one vendor-reported benchmark by under a point: 59.5 vs 58.6 on SWE-Bench Pro. The scores are Meituan's own and unverified, so treat 'beats GPT-5.5' as a claimed ceiling rather than a proven result.

## Sources

- [Meituan open sources LongCat-2.0, the 1.6T near-frontier agentic coding model trained entirely on Chinese chips](https://venturebeat.com/technology/meituan-open-sources-longcat-2-0-the-1-6t-near-frontier-agentic-coding-model-thats-been-leading-openrouter-trained-entirely-on-chinese-chips) — VentureBeat, 2026-06-30
- [meituan-longcat/LongCat-2.0](https://huggingface.co/meituan-longcat/LongCat-2.0) — Hugging Face, 2026-06-29
- [Meituan Open Sources LongCat-2.0 Under MIT License](https://www.opensourceforu.com/2026/06/meituan-open-sources-longcat-2-0-under-mit-license/) — Open Source For You, 2026-06-29