On September 22nd, MEITUAN-W (03690) launched its high-efficiency reasoning model LongCat-Flash-Thinking. According to the company, based on AIME25 test data, LongCat-Flash-Thinking demonstrates superior intelligent agent tool-calling capabilities within this framework, achieving 64.5% token savings compared to non-tool-calling approaches while maintaining 90% accuracy. The model is now fully open-sourced on HuggingFace and Github.

According to official information, this model not only enhances autonomous tool-calling capabilities for intelligent agents but also extends formal theorem proving abilities, making it the first domestic large language model to combine both "deep thinking + tool calling" and "informal + formal" reasoning capabilities. LongCat-Flash-Thinking shows particularly significant advantages in handling ultra-high complexity tasks such as mathematics, coding, and intelligent agent operations.

Comprehensive evaluations indicate that LongCat-Flash-Thinking has achieved state-of-the-art (SOTA) performance among global open-source models across multiple domains including logic, mathematics, coding, and intelligent agent reasoning tasks.