How many tokens does an image use in GPT-4o?

In high detail, GPT-4o scales the image to fit 2048×2048, then scales the shortest side down to 768px, counts 512×512 tiles, and charges 85 + 170 × tiles. A 1024×1024 image becomes 768×768 = four tiles = 85 + 680 = 765 tokens. In low detail it is a flat 85 tokens, regardless of size.

How is image cost calculated in the OpenAI API?

Image inputs are billed as input tokens. You take the token count from the tiling formula and multiply by the model's input rate: cost = tokens ÷ 1,000,000 × rate. For GPT-4o at $2.50 per 1M input tokens, a 765-token image costs $0.0019125 — about Rs 0.58 at 305 LKR per dollar.

How many tokens does Claude charge for an image?

Anthropic estimates image tokens as width × height ÷ 750. A 1024×1024 image is 1,048,576 ÷ 750 ≈ 1,398 tokens. If the longest edge is over 1568px, Claude resizes the image down first, so very large images are capped before the division is applied.

How does Gemini count tokens for images?

Gemini charges 258 tokens for any image that is 384px or smaller on both edges. Larger images are split into 768×768 crops at 258 tokens each, so a 1024×1024 image is a 2×2 grid = four crops = 1,032 tokens. Token count is the same across Gemini Flash and Pro; only the per-token price differs.

Does low-detail mode in GPT-4o reduce image cost?

Yes, significantly. Low detail skips tiling entirely and charges a flat 85 tokens (2,833 on GPT-4o mini) no matter the resolution. That is far cheaper than high detail for large images, at the cost of the model seeing a lower-resolution version. Use low detail when you only need a rough sense of the image.

Why is GPT-4o mini so much more expensive in tokens than GPT-4o?

GPT-4o mini uses the same tiling but reports image tokens at roughly 33.3× the GPT-4o count (base 2,833, per-tile 5,667), so a 1024×1024 image is about 25,501 tokens. Its per-token price is much lower ($0.15 vs $2.50 per 1M), so the dollar cost can still be similar — always compare the cost column, not just tokens.

Is my image uploaded anywhere when I drop it in?

No. When you drop or pick an image, the tool reads only its pixel width and height in your browser using the standard image API, then discards it. Nothing is sent to a server, so you can safely check client work, screenshots, or private photos.

Which exchange rate should I use for the LKR figures?

The calculator defaults to 305 LKR per USD, near the Central Bank of Sri Lanka indicative band in mid-2026, and the field is editable. For billing forecasts, use the rate your card issuer or bank applies to foreign API charges, which usually carries a small margin over the indicative rate.

Are these the real, current provider prices?

The per-1M input-token rates and the tiling constants were last cross-checked against the OpenAI, Anthropic, and Google vision and pricing pages on 2026-06-14. Providers change rates periodically — treat the output as a close estimate and confirm against your latest invoice for exact figures.

AI · Developer

GPT-4o Image Token & Vision Cost Calculator

Enter an image's width and height and see how many tokens it costs on GPT-4o, GPT-4o mini, Claude, and Gemini — plus the price per image and total in USD and LKR, side by side. Each provider's formula is taken from its own docs. The image never leaves your browser.

By Induwara Ashinsana— Executive Director, Ryzera TechnologiesUpdated Jun 14, 2026

Vision token & cost

Image width (px)

The image's pixel width.

Image height (px)

The image's pixel height.

Number of images

How many images you'll send.

OpenAI detail

Affects GPT-4o models only.

Common sizesRead size from image

Models to compare

USD → LKR rate

CBSL indicative rate. Edit to match your bank.

Cheapest total

$0.000103

Gemini 2.0 Flash · Rs 0.03

Tokens / image

1,032

on Gemini 2.0 Flash

Images costed

1,024 × 1,024px each

Model	Tokens/image	Cost/image	Total	Total LKR
Gemini 2.0 Flash Cheapest Google · $0.10/1M in	1,032	$0.000103	$0.000103	Rs 0.03
GPT-4o OpenAI · $2.50/1M in	765	$0.001913	$0.001913	Rs 0.58
GPT-4o mini OpenAI · $0.15/1M in	25,501	$0.003825	$0.003825	Rs 1
Claude Sonnet Anthropic · $3.00/1M in	1,398	$0.004194	$0.004194	Rs 1

OpenAI tiling breakdown

Original: 1,024 × 1,024px
Fit 2048²: 1,024 × 1,024px
Shortest side → 768: 768 × 768px
Tiles: ceil(768/512) × ceil(768/512) = 2 × 2 = 4
Tokens = 85 + 170 × 4 = 765

Token formulas follow each provider's published vision docs. Image inputs bill as input tokens; output/prompt-text tokens are separate.

All math runs in your browser. Dropped images never leave your device.

How it works

A vision model doesn't see your image as pixels — it converts the image into tokens, the same units it charges for text, and bills them at its input-token rate. Every provider counts those tokens with a different, published algorithm, so the same 1024×1024 photo can be 765 tokens on one model and 25,501 on another. This calculator reproduces each algorithm exactly from the source docs.

OpenAI (GPT-4o family, high detail). The image is first scaled to fit inside a 2048×2048 box, then scaled down so its shortest side is 768px. The result is covered with 512×512 tiles — tiles = ceil(w/512) × ceil(h/512) — and tokens are base + per_tile × tiles. GPT-4o uses base 85, per-tile 170; GPT-4o mini reports about 33.3× that (base 2,833, per-tile 5,667). Low detail skips tiling and charges a flat base. The breakdown panel in the tool shows every step.

Anthropic (Claude). Claude estimates image tokens as (width × height) / 750, rounded to the nearest whole token. If the longest edge is above 1568px, Claude resizes the image down proportionally first, so oversize images are capped before the division.

Google (Gemini). An image that is 384px or smaller on both edges is a flat 258 tokens. Anything larger is tiled into 768×768 crops at 258 tokens each: 258 × ceil(w/768) × ceil(h/768). The token count is identical across Gemini Flash and Pro — only the price per token differs.

Cost. For all providers, image inputs are billed as input tokens, so cost = tokens / 1,000,000 × input_rate, then the LKR figure is the USD cost times your exchange rate. Totals multiply by the number of images. Constants and rates are cross-checked on each build; the GPT-4o mini multiplier is verified against OpenAI's documented 33.3× factor so its large token counts are auditable rather than guessed.

Worked examples

1024×1024 image — three providers compared

GPT-4o (high): fit 2048² → 1024×1024; shortest side → 768 ⇒ 768×768
Tiles: ceil(768/512) × ceil(768/512) = 2 × 2 = 4
GPT-4o tokens = 85 + 170 × 4 = 765 → $0.0019125 (Rs 0.58 @305)
Claude Sonnet: round(1024 × 1024 / 750) = 1,398 → $0.004194 (Rs 1.28)
Gemini 2.0 Flash: 258 × (2 × 2) = 1,032 → $0.0001032 (Rs 0.03)
Cheapest by far: Gemini 2.0 Flash

10,000 receipt photos at 1024×1024 (budget check)

GPT-4o: 765 tokens × 10,000 = 7,650,000 tokens → $19.13 (Rs 5,834 @305)
Claude Sonnet: 1,398 × 10,000 = 13,980,000 tokens → $41.94 (Rs 12,792)
Gemini 2.0 Flash: 1,032 × 10,000 = 10,320,000 tokens → $1.03 (Rs 315)
Only Gemini fits comfortably inside a $50 image-processing budget.

Edge case — a 4096×8192 banner on GPT-4o, high detail

Longest edge 8192 > 2048 ⇒ scale ×0.25 → 1024×2048
Shortest side 1024 → 768 ⇒ scale ×0.75 → 768×1536
Tiles: ceil(768/512) × ceil(1536/512) = 2 × 3 = 6
Tokens = 85 + 170 × 6 = 1,105 → $0.0027625 (Rs 0.84 @305)
The 2048 fit + 768 short-side cap stop huge images from exploding in cost.

Frequently asked questions

Sources & references

Algorithms and per-1M input-token rates were last cross-checked against these sources on 2026-06-14. Providers revise vision pricing periodically — confirm against your latest invoice for exact billing.

Related tools

LiveAI

Reasoning Token Cost Calc

Estimate the true cost of reasoning-model API calls by accounting for the hidden reasoning/thinking tokens that o-series, Claude, and Gemini bill at output rates. See per-call and monthly cost in USD and LKR, plus how much more it is than a naive estimate.

Open tool

LiveAI

AI Rate Limit Calculator

Computes whether an LLM workload will hit OpenAI, Anthropic, or Gemini rate limits — effective max requests/min, which limit binds (RPM vs TPM/ITPM/OTPM/RPD), and batch wall-clock time.

Open tool

LiveAI

AI Max Output Tokens

Look up the maximum output (completion) tokens for every current LLM — Claude, GPT-4o, Gemini, Llama and more — and check whether your desired response fits in a single API call or needs chunking. Per-model caps cited from vendor docs, separate from the context window.

Open tool

Rate this tool

Be the first to rate

Comments & feedback

Spotted a bug or want an improvement? Tell us — our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Found a bug, edge case, or want another model added?

Email me at [email protected] — most fixes ship within 24 hours.