Same photo billed 1,120 tokens on Gemini vs 14,364 on GPT-5.6

On October 20, 2026, a DEV Community article compared how different vendors convert images into input tokens for billing. According to the author, the same phone photo is counted as 1,120 tokens on Gemini but 14,364 tokens on GPT-5.6, a difference of more than ten times. The article notes that vision input has no separate price list and is billed as ordinary input tokens, with each vendor using its own formula to turn pixels into tokens, so cost differences can exceed tenfold. The author also mentions having previously written about the cost of generating images or videos, while this post addresses the opposite direction: the picture a user sends into a model so it can read a screenshot, an invoice, or a photo of a whiteboard. As of this report, the matter rests on that comparison; the material does not provide the specific conversion formulas used by each vendor, nor any response from the companies.