The "Research Report on AI Foundation Models and Their Applications in Automotive Field, 2024-2025" from ResearchAndMarkets.com highlights the significant advancements in AI foundation models and their growing impact on the automotive industry. The report emphasizes the key areas of reasoning, cost reduction, and explainability, which are driving the performance and adoption of these models. Reasoning Capabilities Drive Performance Starting from the second half of 2024, AI foundation model companies both in China and globally have been enhancing their models with reasoning capabilities. This has been achieved through reasoning frameworks like Chain-of-Thought (CoT) and its variants such as Tree-of-Thought (ToT), Graph-of-Thought (GoT), and Forest-of-Thought (FoT). These frameworks help AI models handle complex tasks and make independent decisions, which is crucial for applications like autonomous driving and cockpit assistance. In the automotive sector, improved reasoning capabilities have addressed several pain points in AI applications. For instance, they have enhanced the intent recognition of cockpit assistants in complex conversational contexts and improved the accuracy of spatiotemporal predictions in autonomous driving. These advancements are expected to lay the groundwork for more sophisticated Agent applications in the near future. Explainability Enhances User Trust Trust is a critical factor in the widespread adoption of AI systems, especially in the automotive industry. To address this, the report underscores the importance of explainability in AI models. Explainability can be achieved at three levels: data explainability, model explainability, and post-hoc explainability. By demonstrating the reasoning process, AI systems can become more transparent and trustworthy. One notable example is Li Auto, which uses AI reasoning visualization technology in its Level 3 (L3) autonomous driving system. This technology presents the entire decision-making process from physical world perception to driving output, enhancing users' trust in the system. Furthermore, various reasoning models like DeepSeek R1 and Zhipu's GLM-Zero-Preview provide detailed explanations in natural language, breaking down the reasoning process to make it understandable for users. Cost Reduction and Performance Improvement Contrary to the assumption that improved AI capabilities would lead to higher costs, the report highlights that some models, like DeepSeek, have managed to lower the barrier to entry. Early in 2025, Original Equipment Manufacturers (OEMs) began connecting to DeepSeek to enhance the capabilities of their vehicle foundation models. This move was driven by the following factors: Strong Reasoning Performance: DeepSeek's R1 reasoning model is comparable to OpenAI's models and even excels in certain areas like mathematical logic. Lower Costs: DeepSeek maintains high performance while keeping training and reasoning costs low, which is a significant advantage for OEMs. DeepSeek's technologies facilitate cost-effective deployment of high-level autonomous driving and cockpit assistants. For example, low computing overhead technologies allow high-performance models to be deployed on low-compute automotive chips, reducing reliance on expensive GPUs. Techniques such as the DualPipe algorithm and FP8 mixed precision training optimize computing power utilization, making it feasible for mid- and low-end vehicles to support advanced AI features. Impact on Real-Time Performance In driving environments, real-time performance is essential. Autonomous driving systems need to process vast amounts of sensor data quickly, while cockpit assistants must respond swiftly to user commands. Despite the limited computing resources in vehicles, DeepSeek's models have demonstrated enhanced real-time performance. For instance, during server-side training, DeepSeek achieves 90% utilization of NVIDIA A100 chips, reducing inference response times from 20 milliseconds to 9-10 milliseconds on the Qualcomm 8650 platform with 100TOPS of computing power. Key Topics Covered The report provides a comprehensive overview of AI foundation models, their types, and common technologies. It also profiles leading AI foundation model companies and offers detailed application cases in the automotive industry. Additionally, the report explores emerging trends and future applications of these models. Industry Evaluation and Company Profiles Industry insiders commend the advancements in reasoning and explainability as crucial for building trust and driving adoption of AI in the automotive sector. They note that models like DeepSeek are setting a new standard by offering strong performance at lower costs, making AI more accessible to a broader range of vehicles. Li Auto, known for its focus on autonomous driving and user experience, and DeepSeek, recognized for its innovative and cost-effective solutions, are key players in this evolving landscape. Overall, the report suggests that the automotive industry is on the cusp of a significant transformation driven by AI foundation models, with the potential to enhance both driving safety and user convenience. The integration of advanced reasoning and explainability technologies, coupled with cost-effective solutions, is expected to accelerate this transformation and bring smarter, more reliable AI systems to the market.