1、Barclays Capital Inc.and/or one of its affiliates does and seeks to do business with companiescovered in its research reports.As a result,investors should be aware that the firm may have aconflict of interest that could affect the objectivity of this report.Investors should consider thisreport as on

2、ly a single factor in making their investment decision.Please see analyst certifications and important disclosures beginning on page 5.U.S.InternetCharts Worth a Look:OpenAIsDeepSeek Moment Went Over Witha Whimper,But DownloadNumbers Tell Another StoryThe quiet release of OpenAIs open-source AI mode

3、ls is amajor positive development for the Wests AI ecosystem,butmodel performance still lags the state-of-the-art proprietary/closed models.This launch should help continue toaccelerate AI development while bending the cost curvedown.Our Take:Last week,OpenAI(private,not covered)released two new ope

4、n-sourced largelanguage models,the first time the company has open-sourced an LLM since the release ofGPT-2 in 2019.The release of these new models,GPT-OSS-120B and GPT-OSS-20B,was separatefrom OpenAIs release of the GPT-5 model(also last week).In our view,OpenAIs big stepforward in open-sourced mod

5、el capabilities marks an important milestone in the developmentof the Wests AI ecosystem.Interestingly,the$4m training cost(likely final run)is lower thanDeepSeeks V3 or R1 albeit a smaller base model,and performance is better,so Western labs dohave similar technical chops around efficiency when tas

6、ked to optimize for that.Quick Background on Open-Sourced LLMsWhen an AI lab open-sources a large language model,this generally means that the modelweights are openly released to the public(open-weight).Model weights are the values thatthe LLM learns during the training process,determining how the m