JOHNISKING/image-token-cost

Image input token cost by resolution: Claude vs Gemini vs GPT

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)
• DOI (Zenodo): 10.5281/zenodo.23443906
• 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

1. 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.
2. 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.
3. Gemini: per-image token count shown by Google AI Studio for Gemini 3.8 Flash at the default media resolution.
4. 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.

Citation

TokenSave (2026). Image input token cost by resolution: Claude vs Gemini vs GPT [Data set]. Zenodo. https://doi.org/10.5281/zenodo.23443906

@misc{tokensave2026image,
  title  = {Image input token cost by resolution: Claude vs Gemini vs GPT},
  author = {TokenSave},
  year   = {2026},
  publisher = {Zenodo},
  doi    = {10.5281/zenodo.23443906},
  url    = {https://tokensave.app/data/image-token-cost},
  note   = {Data set, CC BY 4.0}
}

Count tokens for your own text at tokensave.app.