
Mind GPT
AI Lixiang Tongxue, growing alongside the family.
Mind GPT, Li Auto's self-developed large multimodal cognitive model. Powered by Mind GPT, Lixiang Tongxue can answer any question from family members. Whether it's about driving, traveling, entertainment or more, Lixiang Tongxue is always there, understanding your words, answering your questions,and helping you plan and strategize.

Mind GPT truly grasps the needs of your family, offering fresh perspectives and inspiration for their creations. Whether it's composing poetry, painting, taking photos, or telling stories, just a sentence is all it takes to assist you. Your family's AI assistant, there to support and unleash creativity every step of the way.

Lixiang Tongxue has developed a "Children Mode" tailored for children at home, providing more child-friendly content and serving as an all-around companion for every child's journey into the broader world.

What makes Mind GPT so impressive?
Multimodal human-machine interaction technology.
Model Capabilities
Text Generation
Through learning and training on vast amounts of linguistic data, Mind GPT can automatically generate text content such as articles, conversations, etc.
Text Generation
Through learning and training on vast amounts of linguistic data, Mind GPT can automatically generate text content such as articles, conversations, etc.
Image Generation
With powerful intelligent algorithms and image generation capabilities, without the need for complex design or drawing skills, unique images can be created with just a single sentence. It turns everyone into a great artist in seconds.
Image Generation
With powerful intelligent algorithms and image generation capabilities, without the need for complex design or drawing skills, unique images can be created with just a single sentence. It turns everyone into a great artist in seconds.
External Information Retrieval
By connecting to external Model Zoo and API Zoo, while dispatching external APIs and dedicated models, Lixiang Tongxue can access to external knowledge in real-time.
External Information Retrieval
By connecting to external Model Zoo and API Zoo, while dispatching external APIs and dedicated models, Lixiang Tongxue can access to external knowledge in real-time.
Memory Network
Lixiang Tongxue comes with a built-in memory network that can accurately store a vast amount of knowledge, while iterating based on user queries, commands, and feedbacks, which helps it understand each family member better.
Memory Network
Lixiang Tongxue comes with a built-in memory network that can accurately store a vast amount of knowledge, while iterating based on user queries, commands, and feedbacks, which helps it understand each family member better.
Authoritative Evaluation
Champion of the C-Eval Comprehensive Evaluation for Chinese Language Models
C-Eval stands as a comprehensive examination suite for Chinese LLMs, co-developed by Tsinghua University, Shanghai Jiao Tong University, and the University of Edinburgh. Mind GPT clinched the top spot among the large language models participating in the evaluation.
Champion of the C-Eval Comprehensive Evaluation for Chinese Language Models
*Data as of December 15, 2023
CMMLU Chinese Large Model Comprehensive Evaluation Benchmark Champion
CMMLU is a Chinese Multi-Task Language Understanding benchmark constructed by MBZUAI, Shanghai Jiao Tong University, Microsoft Research Lab - Asia, and the University of Melbourne.Mind GPT ranks first among the large language models participating in the evaluation.
CMMLU Chinese Large Model Comprehensive Evaluation Benchmark Champion
*Data as of December 15, 2023. The CMMLU benchmark includes two sub-benchmarks, Five-shot and Zero-shot, representing different evaluation methods. The Five-shot sub-benchmark assesses large language models by providing five specified examples as references, while the Zero-shot sub-benchmark evaluates large language models without any input examples as references.
Technical Support
Crystal Clear Vision
RGB+IR Visual Module
rand-new cabin RGB, IR perception hardware, and 3D vision algorithms, not only provide Lixiang Tongxue with clear, comprehensive, and understandable vision during the day but also equip it with the ability to respond interactively at night.*The function is locally implemented on the automotive head unit system and triggered by user commands. Specific configuration
Crystal Clear Vision
rand-new cabin RGB, IR perception hardware, and 3D vision algorithms, not only provide Lixiang Tongxue with clear, comprehensive, and understandable vision during the day but also equip it with the ability to respond interactively at night.*The function is locally implemented on the automotive head unit system and triggered by user commands. Specific configuration
Audible
Dual-array microphones
By employing multiple sets of Dual-array Microphone technology throughout the vehicle, we achieve more precise spatial sound source localization, enabling more accurate personalized speech recognition across multiple languages, dialects, and tasks.
Audible
By employing multiple sets of Dual-array Microphone technology throughout the vehicle, we achieve more precise spatial sound source localization, enabling more accurate personalized speech recognition across multiple languages, dialects, and tasks.
Powerful Computing Power
High-performance Automotive Head Unit Chip
Powerful Computing Power
The Qualcomm Snapdragon 8295P high-performance head unit chip, coupled with a 5G dual-card dual-standby high-speed communication module, empower Lixiang Tongxue equipped with Mind GPT with faster response time and enhanced comprehension ability.
Superintelligence
In-house Developed Large Cognitive Model
Superintelligence
The Model utilizes the proprietary TaskFormer neural network architecture for training, boasting over 111 domains and 1000+ exclusive capabilities. These encompass various scenarios such as family transportation, entertainment, and travel, making it a comprehensive model that better understands automobiles and families.