Meituan ships LongCat-2.5-Preview: 1.6T MoE multimodal agent with 1M context

Meituan launched LongCat-2.5-Preview on its LongCat API platform, a long-horizon multimodal agent model built on a mixture-of-experts architecture with about 1.6 trillion total parameters and roughly 48 billion active per inference step. It offers a native 1M-token context window, 128K max output, and is available via API and at longcat.chat.

• MoE architecture: ~1.6T total parameters, ~48B active per inference step
• Native 1M-token context window, 128K max output length
• Native multimodal image understanding built into the base model
• Promo pricing: ~$0.30 input and ~$1.20 output per million tokens