The Chinese company’s newest multimodal model has been made available for use inside developer tools such as Claude Code, Codex, and Hermes to assist with high-complexity software projects. To support these agentic tasks, LongCat-2.5 now features a 1M-token context window and is offered at promotional API rates of $0.30 per million uncached input tokens and $1.20 per million output tokens.
The system utilizes a mixture-of-experts architecture with 1.6T total parameters, of which approximately 48B are active during operations. Designed for long-horizon tasks including 3D creative design and visual coding, the model is now available to developers via APIs and is positioned as a competitor to other large-scale systems like GPT-5.5 and Claude-Opus-5.
