汇丰：中国软件行业研究：DeepSeek-R1引领AI商业化进阶-250205（中译版）(11页）.pdf

1、 Disclosures&Disclaimer This report must be read with the disclosures and the analyst certifications in the Disclosure appendix,and with the Disclaimer,which forms part of it.Issuer of report:HSBC Qianhai Securities Limited View HSBC Qianhai Securities at:https:/ Equity Research Report DeepSeek-R1 s

2、trong performance at low cost:DeepSeek-R1 is a reasoning model developed by DeepSeek,a Chinese AI startup.As a reasoning model(vs general model),it first generates a Chain of Thought to enhance the accuracy of responses before delivering the final answer.This open-source model has shown robust perfo

3、rmance(rivaling OpenAIs o1 in mainstream benchmarks)at a much lower cost in model training and API(output cost 96%cheaper than o1),mainly owing to R1s engineering innovation.DeepSeek-R1 has significantly lowered thresholds for app developers and has enabled complex applications with high reasoning r

4、equirements.Implications for AI application monetization:In contrast with Bytedances LLM Doubao(豆包),whose expertise is content generation and multimodal,DeepSeek-R1 stands out in logical reasoning.Before R1,domestic LLMs logic capabilities were insufficient for productivity AI tools,and AI applicati

5、on monetization progress had disappointed the market.In our view,R1s performance and very low token costs should significantly cut AI training/inference costs,help AI technology expand,and bring in diversified AI applications.Meanwhile,the adoption of R1 could improve software companies R&D efficien

6、cy,saving costs and driving earnings growth,medium term.Focus on AI agents:The monetization of AI agents in US companies has begun.Besides OpenAI Operator,the intelligent individual secretary,several software companies have launched AI agents in subsectors for instance,Salesforces Agentforce and Ser