Image input token cost by resolution: Claude vs Gemini vs GPT
Abstract
Input tokens for one image at 10 resolutions (256×256 to a 12 MP photo) on Claude Opus/Sonnet 5.5, Claude Haiku 4.5, Gemini 3.x and GPT. Headline: a 12-megapixel phone photo costs 1,065 input tokens on Gemini, 4,005 on Claude 5.5 and 14,364 on GPT-6 Astra with default settings. Gemini charges about the same for every image, so it is the cheapest for large pictures and the most expensive for small ones. Source page, chart and method: https://tokensave.app/data/image-token-cost Related: token cost by language (https://tokensave.app/data/token-cost-by-language-claude-gemini) and by data format (https://tokensave.app/data/token-cost-by-format) Measured: 2026-10-18 License: CC BY 4.0 Key findings Gemini charges a flat rate: 1,065 to 1,101 tokens per image, from a 256×256 icon to a 12 MP photo. For the photo that is 93% less than GPT-6 Astra. GPT-6 Astra keeps the original size by default, so a 12 MP photo costs 14,364 tokens. detail: "high" caps it at 2,942 (-80%). Claude 5.5 and Haiku 4.5 cost almost the same up to about 1 MP. Above that Claude 5.5 needs 70% more for a 1080p screenshot and 141% more for a 12 MP photo. Shrinking a 12 MP photo to 1024×1024 cuts its cost by 64% on Claude 5.5 and 91% on GPT-6 Astra. Files image-token-cost.csv: one row per image size with size, width, height, megapixels and tokens for claude55, claude45, gemini, gpt6_auto, gpt6_high, gpt55_auto image-token-cost.json: the same data plus column definitions Data Image size Claude Opus/Sonnet 5.5 Claude Haiku 4.5 Gemini 3.x GPT-6 Astra (default) GPT-6 Astra (detail: high) GPT-5.5 (default) 256×256 177 165 1,090 77 77 77 512×512 438 426 1,090 308 308 308 768×768 861 849 1,090 692 692 692 1024×1024 1,446 1,434 1,090 1,229 1,229 1,229 1280×720 1,273 1,261 1,101 1,104 1,104 1,104 1920×1080 2,768 1,625 1,101 2,448 2,448 2,448 1080×1920 2,768 1,625 1,101 2,448 2,448 2,448 2560×1440 3,069 1,657 1,101 4,320 2,931 4,320 3840×2160 3,069 1,657 1,101 9,792 2,931 9,792 4032×3024 4,005 1,661 1,065 14,364 2,942 11,868 Method One synthetic PNG rendered at 10 sizes. A second image with different content gave identical counts on Claude 5.5 and Gemini, so counts depend on size, not content. Claude: each image sent through Claude Code (claude -p, stream-json input) with a fixed system prompt and a one-character message; input tokens minus the same run without the image. Opus 5.5 and Sonnet 5.5 gave identical counts. Includes the small per-image message overhead. Gemini: per-image token count shown by Google AI Studio for Gemini 3.8 Flash at the default media resolution. GPT: computed with the 32×32-pixel patch formula and model limits in OpenAI's vision guide (patch count × 1.2); reproduces the guide's worked examples. Not measured through the API. Limitations Input tokens only; price per token differs by model. Default settings; Gemini media resolution and GPT detail change the counts. Claude counts are from Claude Code; the API may differ by a few tokens of message overhead.
