计算机：Deepseek模型发布，提振国内AI发展

Investment Rating - The industry investment rating is "Positive," indicating an expectation that the industry index will outperform the market index by more than 5% over the next six months [5]. Core Insights - The release of the DeepSeek-R1 model by DeepSeek, a subsidiary of Huansheng Quantitative, significantly boosts domestic AI development, showcasing performance comparable to OpenAI's o1 model in various tasks [1]. - DeepSeek's R1 model employs reinforcement learning techniques during the post-training phase, enhancing its reasoning capabilities with minimal labeled data [1]. - The API pricing for DeepSeek's model is notably competitive, with costs of 1 RMB per million input tokens (cache hit) and 4 RMB (cache miss), and 16 RMB per million output tokens, which is 27-55 times lower than OpenAI's o1 model [1]. Summary by Sections Event - On January 20, DeepSeek officially launched the DeepSeek-R1 model, which performs exceptionally well in mathematical, coding, and natural language reasoning tasks [1]. Model Performance - The R1 model's performance is on par with OpenAI's o1 official version, indicating a significant advancement in AI capabilities [1]. Training Techniques - The model's training involved extensive use of reinforcement learning, which greatly improved its inference abilities despite limited labeled data [1]. API Pricing - DeepSeek offers a competitive pricing structure for its API, making it an attractive option for users compared to existing models in the market [1].