Li Auto officially announced a major upgrade to its self-developed MindGPT large model, with the new version MindGPT 3.1 making its debut. This upgrade deeply integrates the capabilities of intelligent agents into the large model, enabling the innovative "think and search" function, which allows the model to simultaneously call various tools during reasoning, providing users with faster, more comprehensive, and accurate results.
According to official data, MindGPT 3.1 has achieved a qualitative leap in performance, with a maximum output speed of up to 200 tokens per second. Compared to the previous version, MindGPT 3.0, the speed has increased nearly five times. This breakthrough will greatly enhance the interaction efficiency between users and the smart assistant.
MindGPT 3.1 not only impresses with its speed but also demonstrates exceptional intelligence in interactive capabilities. It deeply integrates reasoning chain thinking and tool calling capabilities. The model can break down complex problems step by step, like an experienced analyst, through a cycle mechanism of "self-thinking - self-tool calling - further reasoning," continuously optimizing answer quality and significantly improving the success rate of completing complex tasks.
In terms of core capabilities, MindGPT 3.1 also performs outstandingly. Its indicators are comprehensively better than those of MindGPT 3.0 and outperform industry open-source models such as Qwen3-235B in areas like mathematics, code, science questions, and instruction following. Especially in code capabilities, MindGPT 3.1 has been further enhanced, capable of easily implementing classic programming examples such as Snake and Pong control, demonstrating its strong technical power.
