SemiAnalysis, an industry analysis firm, recently carried out practical tests on the multi-tier subscription services offered by OpenAI and Anthropic. Throughout these tests, the firm acquired a range of subscriptions from both entities and persistently engaged in resource-intensive activities, including long-duration programming and 'agents' operations, until they hit the weekly usage caps. Following this, they computed the theoretical costs associated with this level of usage, drawing on publicly accessible API pricing information. The findings unveiled a substantial disparity in computational power subsidies, lurking beneath the deceptively reasonable monthly subscription fees.
