Meituan, the Beijing-based delivery and local services company, said on July 6 that it open-sourced LongCat-2.0.

Chinese GPU startup Moore Threads said the same day that it had adapted the model for inference on its MTT S5000 AI training-and-inference GPU card using the MUSA software stack.

The adaptation covered model loading, inference engine launch, operator optimization, deployment verification, and accuracy validation.

The company did not disclose any performance data.

The release is part of Meituan’s wider AI efforts.

The company has invested in AI and domestic chips as it seeks to expand beyond logistics and local services.

Meituan has deployed LongCat-based models in its app to support customer service and multi-agent interaction systems.

The company said the model family already supports its core local-services operations for users, merchants, and riders.

🔗 Source: Yicai

Recent Meituan developments