On January 16, Meituan announced that LongCat-Flash-Thinking-2601, an upgraded version of the LongCat-Flash-Thinking model, is now open-source.
The new model has achieved State-of-the-Art (SOTA) performance on key evaluation benchmarks, including agent search, agent tool calling, and tool interaction reasoning. In random complex tasks that rely on tool calling, its performance surpasses that of Claude-Opus-4.5-Thinking, significantly reducing the adaptation training costs for new tools in real-world scenarios. The model also supports a "deep-thinking" mode, capable of activating eight parallel "brains" to execute tasks, which can be experienced for free on the LongCat official website.
Comments