This article details the quota consumption of the Gemini 2.5 Flash and Gemini 3 Pro (Low) models through continuous testing of the Google Gemini API. The tests showed that both models hit their quota limits simultaneously after the 17th conversation, with identical reset times. Based on this, the author speculates that the High and Low versions of Gemini 3 Pro may have no practical difference, with all requests potentially being routed to the same Low-tier service. The article also analyzes the patterns of quota consumption, pointing out that the officially advertised ‘relaxed rate limiting’ actually has usage restrictions within time windows, and the retry mechanism is confusing when frequent errors occur. This analysis provides valuable insights for developers and researchers to understand Google Gemini model quota limits and usage strategies, and also serves as a case study for evaluating the transparency of AI model service providers.
Original Link:Linux.do


评论前必须登录!
立即登录 注册