model-release coding-with-ai developer-tools industry ai-tools

China's Meituan Open-Sources 1.6 Trillion Parameter Coding Model

Meituan releases LongCat-2.0 under MIT license — a 1.6T parameter model trained entirely on Chinese chips that outperforms GPT-5.5 on coding.

AI Tutorials · 30 June 2026 · 2 min read

Quick answer

Meituan has open-sourced LongCat-2.0, a 1.6 trillion parameter coding model trained entirely on domestic Chinese chips. It uses a Mixture-of-Experts architecture, supports a 1 million token context window, and scores 59.5 on SWE-bench Pro — beating GPT-5.5's 58.6. The model is available now on Hugging Face under the MIT license.

product-launch

China's Meituan Open-Sources 1.6 Trillion Parameter Coding Model

Chinese tech giant Meituan has open-sourced LongCat-2.0, a 1.6 trillion parameter AI model designed for coding — and it was trained entirely on domestically produced Chinese chips. The model is available now on Hugging Face under the MIT license, making it one of the largest open-source models ever released.

What Makes LongCat-2.0 Notable

LongCat-2.0 uses a Mixture-of-Experts (MoE) architecture that activates roughly 48 billion parameters per token out of its 1.6 trillion total. This makes it far more efficient than running a dense model of comparable size. It supports a 1 million token context window — enough to process an entire large codebase in a single session.

On coding benchmarks, the model punches well above expectations. It scored 59.5 on SWE-bench Pro, surpassing OpenAI’s GPT-5.5 at 58.6. It also posted 70.8 on Terminal-Bench 2.1 and 77.3 on SWE-bench Multilingual. These are near-frontier results from a fully open-source model.

The Hardware Story

What makes this release geopolitically significant is the training infrastructure. Meituan built LongCat-2.0 on a cluster of 50,000 domestically produced AI chips — reportedly Huawei Ascend processors — without any Nvidia or other restricted Western hardware. The team trained across more than 35 trillion tokens, navigating challenges in communication faults, memory pressure, and numerical stability at massive scale.

This directly challenges assumptions about US export controls. The restrictions were designed to slow China’s AI progress by limiting access to advanced chips, but LongCat-2.0 suggests that gap is narrowing faster than expected.

What This Means for You

If you work with AI coding tools, LongCat-2.0 is worth watching. As an MIT-licensed model, it can be used commercially without restriction. It is already available through API providers like OpenRouter, and its 1 million token context makes it well-suited for agentic coding workflows where an AI needs to understand an entire project at once.

The broader takeaway: the open-source AI ecosystem continues to close the gap with proprietary models. Between LongCat-2.0, Qwen 3.5, and other recent releases, developers now have genuinely competitive alternatives to closed-source tools from OpenAI and Google.

Frequently asked questions

What is LongCat-2.0?

LongCat-2.0 is a 1.6 trillion parameter open-source AI model built by Meituan, China's largest food delivery and services platform. It uses a Mixture-of-Experts architecture designed specifically for agentic coding tasks, with a 1 million token context window.

How does LongCat-2.0 compare to GPT-5.5 and other models?

LongCat-2.0 scored 59.5 on SWE-bench Pro, surpassing GPT-5.5 at 58.6. It also achieved 70.8 on Terminal-Bench 2.1 and 77.3 on SWE-bench Multilingual. It is competitive with frontier models on coding tasks while being fully open-source.

What chips was LongCat-2.0 trained on?

LongCat-2.0 was trained on a cluster of 50,000 domestically produced Chinese AI chips, reportedly Huawei Ascend processors. No Nvidia or other Western chips were used, marking a milestone for China's AI self-sufficiency.

Can I use LongCat-2.0 for free?

Yes. LongCat-2.0 is released under the MIT license and available on Hugging Face. You can download the model weights and run it locally, or access it through API providers like OpenRouter.

Why does this matter for the AI industry?

It demonstrates that frontier-quality AI models can be built without access to the most advanced Western chips. This challenges assumptions about the effectiveness of US export controls on AI hardware to China.