# Meituan's LongCat-2.0: the first trillion-parameter model built without Nvidia

> Meituan open-sourced LongCat-2.0 in late June 2026: a 1.6-trillion-parameter MoE trained end-to-end on Chinese chips.

*A 1.6-trillion-parameter open model is the smaller story. The compute stack underneath it — 50,000 domestic chips, no Nvidia — is the one that changes the map.*

By WireRead Editorial · WireRead
Canonical: https://wireread.com/news/meituan-longcat-2-open-source

The number Meituan wants you to read is 1.6 trillion parameters. The number that actually reshapes the competitive map is 50,000 — the count of domestic ASICs in the cluster that trained and now serves LongCat-2.0, with, per Meituan, not a single Nvidia GPU in the loop. If that claim holds up to independent scrutiny, LongCat-2.0 is the first trillion-parameter model built end-to-end on Chinese silicon, and that is a supply-chain fact with far longer legs than any benchmark row.

## What Meituan actually shipped

In late June 2026 Meituan — China's largest food-delivery and local-services platform, not a name most Western developers associate with frontier models — open-sourced LongCat-2.0 under a permissive **MIT licence**. The core specifications:

| Attribute | LongCat-2.0 |
| --- | --- |
| Total parameters | 1.6 trillion |
| Active parameters / token | ~33-56 billion (dynamic) |
| Architecture | Mixture-of-Experts |
| Context window | 1 million tokens (native) |
| Licence | MIT (commercial use permitted) |
| Training/serving hardware | >50,000 domestic ASICs, no Nvidia |

The dynamic activation band — the model routes each token through somewhere between roughly 33 and 56 billion parameters rather than a fixed slice — is the efficiency lever that makes a 1.6T-parameter model economical to run. One caveat at launch: Meituan listed the full weights as 'coming soon' rather than all posted, so the open-weight claim was, at announcement, partly a promise.

> Meituan open-sourced LongCat-2.0 as a 1.6-trillion-parameter Mixture-of-Experts model with a native one-million-token context, and confirmed it was the same system that had been running anonymously as 'Owl Alpha' at the top of OpenRouter's developer usage charts.
> — [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 compute-sovereignty milestone

For two years the working assumption in Western AI policy has been that US export controls on advanced Nvidia accelerators impose a hard ceiling on how large a model China can train. LongCat-2.0 is the first concrete counter-example at the trillion-parameter scale. Meituan's claim is not that domestic ASICs match an H100 chip-for-chip — it is that a cluster of more than 50,000 of them, engineered around, was sufficient to train and serve a near-frontier model end-to-end. That reframes the export-control question from 'can they train a big model?' to 'how much does the domestic-silicon detour actually cost them in time and efficiency?' — a narrower and more uncomfortable question.

> Meituan released LongCat-2.0 under an MIT licence and presented it as the first trillion-parameter model trained end-to-end on Chinese-made chips, framing the achievement as a milestone in domestic compute independence.
> — [Open Source For You](https://www.opensourceforu.com/2026/06/meituan-open-sources-longcat-2-0-under-mit-license/), 2026-06-29

> **Key:** **The throughline.** The story that dates well here is not whether LongCat-2.0 edges GPT-5.5 on one coding test. It is that a Chinese consumer-internet company trained a trillion-parameter model without Nvidia and gave the weights away under MIT. Capability parity is contested and moves monthly; a working non-Nvidia training pipeline at this scale is structural.

## The benchmarks — read them carefully

Meituan positions LongCat-2.0 as a near-frontier agentic coding model, and the figures it published are strong — but every one of them is **vendor self-reported and not yet independently verified**:

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

The widely repeated 'beats GPT-5.5' line is technically true and materially thin: it rests on a single benchmark, SWE-Bench Pro, by a margin of under one point (59.5 vs 58.6) — inside the noise band of most eval harnesses. Treat 59.5 as a claimed ceiling under favourable conditions, not a settled ranking. Independent scorers such as SWE-bench maintainers, Artificial Analysis or LMArena had not published verified numbers at announcement.

## Why the open-weight framing matters

LongCat-2.0 does not arrive in isolation. It is the latest and heaviest entry in a run of Chinese open-weight releases — DeepSeek, Moonshot, MiniMax and now Meituan — that have quietly captured the demand side of the market: Chinese open-weight models now account for roughly **61% of OpenRouter's top-10 traffic**. The 'Owl Alpha' episode is the tell. A model can lead a developer usage chart for two months entirely on merit and price before anyone knows who built it, because open weights and an OpenRouter endpoint remove the brand from the buying decision. That is the mechanism by which the open-weight surge converts into installed base — and installed base, not a leaderboard crown, is what compounds. The variables worth tracking now: whether the full weights ship as promised, whether independent benchmarks confirm the coding scores, and whether anyone outside Meituan can reproduce the non-Nvidia training claim.

## Key takeaways

- Meituan open-sourced LongCat-2.0 in late June 2026 under a permissive MIT licence: a 1.6-trillion-parameter Mixture-of-Experts model with a native 1M-token context, activating roughly 33-56 billion parameters per token.
- The headline claim is a compute-sovereignty milestone: Meituan says LongCat-2.0 is 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 unmasked the model as 'Owl Alpha', the anonymous stealth entry that had led OpenRouter's developer usage charts for around two months before the reveal.
- Meituan-reported benchmarks — 59.5 SWE-Bench Pro (vs GPT-5.5's 58.6), 70.8 Terminal-Bench 2.1, 77.3 SWE-Bench Multilingual — are vendor self-reported and unverified; the 'beats GPT-5.5' claim holds on one test by under a point.
- The release deepens the open-weight surge: Chinese open-weight models now make up roughly 61% of OpenRouter's top-10 traffic.

## FAQ

### What is LongCat-2.0 and who made it?
LongCat-2.0 is a 1.6-trillion-parameter open-weight Mixture-of-Experts language model open-sourced by Meituan — China's largest food-delivery and local-services platform — in late June 2026 under an MIT licence. It has a native 1M-token context and activates roughly 33-56 billion parameters per token.

### Was LongCat-2.0 really trained without Nvidia chips?
Meituan says yes — 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. That is Meituan's claim and had not been independently verified at announcement, but it is the release's central significance if it holds.

### Does LongCat-2.0 actually beat GPT-5.5?
Only narrowly and only on one vendor-reported benchmark. Meituan's own figures put it at 59.5 on SWE-Bench Pro versus GPT-5.5's 58.6 — a margin of under one point, inside typical evaluation noise. The scores are self-reported and unverified, so 'beats GPT-5.5' is a claimed ceiling, not a settled ranking.

### What was 'Owl Alpha'?
Owl Alpha was an anonymous stealth model that topped OpenRouter's developer usage charts for around two months. Meituan revealed at launch that Owl Alpha was LongCat-2.0 running incognito — a sign the model won real developer demand before its origin was known.

### Is LongCat-2.0 free to use?
It is released under an MIT licence, which permits free commercial use and self-hosting. At announcement Meituan listed the full weights as 'coming soon' rather than all posted, so availability was still rolling out — worth checking the Hugging Face repository before planning a deployment.

## 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