OpenAI / ChatGPT

OpenAI image token calculator

Estimate how many input tokens an image costs on OpenAI / ChatGPT (GPT-5 family), and whether packaging your prompt as an image beats sending it as text.

OpenAI's current GPT-5-family vision models bill images at roughly one token per 32×32-pixel patch, rounded up per side. A 500×500 image works out to about 16 × 16 = 256 tokens, so a prompt image only saves tokens once the text it replaces exceeds roughly 256 tokens (~1,000 characters).
Model familyImage tokens (est.)vs 2,000 text tokens
OpenAI~25687% fewer tokens

Formulas: OpenAI — ≈ one token per 32×32-pixel patch (rounded up per side). Estimates only, before any wrapper instructions or caching — verify with the provider's token-counting API. See our methodology.

Common sizes on OpenAI

Square imageEstimated tokensBeats text longer than
384×384~144~144 text tokens
500×500~256~256 text tokens
512×512~256~256 text tokens
768×768~576~576 text tokens
1024×1024~1,024~1,024 text tokens

Notes and caveats

Older tile-based OpenAI families (GPT-4o era) use a different calculation — a base cost plus a per-512-pixel-tile cost that depends on the detail setting — so the same image can be billed very differently across model versions. Always check the model you actually call.

OpenAI also resizes large images before processing. Oversized canvases may be downscaled, which reduces both the token count and the legibility of small text — cheaper is not automatically more readable.

These are input-side estimates before wrapper instructions, caching, latency, retries and accuracy effects. Token rules are model-version-specific and change — verify with the provider's token-counting endpoint, and read when not to convert prompts to images before shipping this in production.

Try it on a real prompt

The visual prompt compressor gives you a live OpenAI-specific estimate, flags exact identifiers that shouldn't go into a lossy image, and exports the PNG plus the original .txt — all locally in your browser.

Last updated July 15, 2026 · Estimates are derived from provider documentation and can change — see methodology.