openai actually released some figures on power use per prompt, but the caveat is that a single prompt to their services can trigger multiple responses (the “thinking” mode) so they’re not consistent.