OpenAI's only public per-query figures come from a June 2025 blog post by CEO Sam Altman citing 0.34 Wh of energy and 0.000085 gallons of water for an 'average query.' No model is specified, no methodology is provided, no audit exists, and no measurement boundary is defined. OpenAI has not published a methodology-backed per-query figure since GPT-3 in 2020.
None disclosed. The numbers appear in a personal blog post without supporting documentation, technical paper, or third-party verification.
GPT-5 and the o-series reasoning models are not covered at all. Independent estimates from Epoch AI and the University of Rhode Island AI Lab place GPT-5 at roughly 18 Wh per medium response, with o3 substantially higher — orders of magnitude above the blog-cited average. The gap between the marketing claim and independent estimates is the central transparency problem.
A single CEO-stated number with no model, no methodology, and no audit is treated as a marketing claim, not a disclosure. Earns partial credit only for the existence of a quantitative statement.