# What an image actually costs, and why it moved when you changed model > Claude charges no per-image fee. An image costs ceil(w/28) x ceil(h/28) visual tokens, and the same 4K image costs 3.07 times more on newer models. URL: https://getculpa.com/image-token-cost Last reviewed: 2026-08-05 Rates effective: 2026-07-02 ## Answer An image is billed as input tokens rather than a flat per-image fee. Claude converts it into visual tokens on a published formula, and newer models allow three times more of them for the same picture. Culpa, a local-first LLM cost, margin, and forecast ledger, prices every call from the tokens the provider reports, images included. ## Why this happens The first surprise is that there's no image line to look for. Providers convert a picture into input tokens and charge the ordinary input rate, so a vision call looks like a text call with a large prompt. Anthropic publishes the conversion: an image is divided into 28-pixel patches and each patch is one visual token, so the count is the width in patches times the height in patches. That makes image cost a function of resolution, which is the second surprise, because resolution is usually set by whatever produced the file rather than by anyone thinking about money. The third is the one that moves bills. Models sit in one of two resolution tiers, and the newer tier permits roughly three times the visual tokens for the same image before downscaling kicks in. Upgrade the model and your image costs rise sharply with no change to your code, your prompts or your pictures. ## What this usually looks like - Your bill rose after a model upgrade and the prompts didn't change. - You looked for a per-image charge and couldn't find one. - Screenshots go to the model at whatever size the screen happened to be. - Nobody knows the average resolution of the images you send. - A vision feature costs more than a text feature doing similar work. ## Common mistakes - Looking for a per-image price. Why it hurts: There isn't one on the models most people use, so the search ends with the wrong conclusion that images are free. Do instead: Convert resolution to visual tokens, then price them at the input rate. - Sending images at whatever size they arrive. Why it hurts: Cost scales with pixels, so an unresized screenshot can cost several times a resized one. Do instead: Downscale before sending, to the smallest size the task still works at. - Assuming a model upgrade keeps image costs flat. Why it hurts: The newer resolution tier allows about three times the visual tokens for the same image. Do instead: Re-price your image workload against the new tier before you migrate. - Treating image cost as fixed per call. Why it hurts: Two calls with one image each can differ by an order of magnitude on resolution alone. Do instead: Record the token count per call, which already carries the image cost inside it. ## Self-check - Take a typical image you send and work out its visual tokens from its dimensions. - Multiply by your model's input rate and by your monthly image volume. - Check which resolution tier your model sits in, and what the other tier would cost. - Find the largest image your product will accept and price that as your worst case. ## The same picture, priced three ways Anthropic's vision documentation, read 2026-08-05, publishes the formula, a token table and worked per-thousand figures. A 1000x1000 image costs ceil(1000/28) x ceil(1000/28) visual tokens. A 4K image at 3840x2160 downscales to 1,560 visual tokens on the standard tier and 4,784 on the high-resolution tier, which covers Claude 4.7 and later. Rates are the price book's real $1.00 per million for Haiku 4.5 and $5.00 for Opus 5. Every figure below is published rather than assumed. 1000x1000: ceil(1000/28) = 36, so 36 x 36 = 1,296 visual tokens 1,296 x $1.00/M x 1,000 images = $1.30 per thousand on Haiku 4.5 1,296 x $5.00/M x 1,000 images = $6.48 per thousand on Opus 5 a 4K image on the high-resolution tier: 4,784 x $5.00/M x 1,000 = $23.92 per thousand the same 4K image on the standard tier is 1,560 tokens, so the tier alone is 4,784 / 1,560 = 3.07x Anthropic states all three dollar figures in its own prose, and re-deriving them from the formula reproduces each one exactly. That's worth noticing: the arithmetic here isn't an interpretation of the pricing page, it agrees with it to the cent. The 3.07x tier difference is the part that arrives unannounced, because a model upgrade changes it without touching a line of your code. ## Cost figures Every figure carries its confidence and its source. For each provider-reported figure below, its own source line names the published record it came from, the date that record was read, and any aggregation applied to it. - $1.30, $6.48 and $23.92 per thousand images, published cost of a 1000x1000 image on two models, and a 4K image on the high-resolution tier [provider-reported] Source: Anthropic's vision documentation at platform.claude.com/docs/en/build-with-claude/vision, read 2026-08-05, states each figure in its own prose. The same page publishes the formula and the token table, and re-deriving from them reproduces all three exactly: 1,296 visual tokens at $1.00 and $5.00 per million, and 4,784 tokens at $5.00 per million, each times 1,000 images. Rates match this price book's Haiku 4.5 and Opus 5 rows. - 3.07x, visual-token cost of one 4K image on the high-resolution tier against the standard tier [calculated] Source: Anthropic's published token table gives 4,784 visual tokens for a 3840x2160 image on the high-resolution tier and 1,560 on the standard tier. 4,784 / 1,560 = 3.07, matching the same page's statement that high-resolution images can use roughly three times more visual tokens. Both token counts are published. ## FAQ Q: How much does an image cost in the Claude API? A: There's no per-image fee. An image becomes input tokens on the formula ceil(width/28) x ceil(height/28), and those tokens bill at the model's normal input rate. A 1000x1000 image is 1,296 visual tokens, which is $1.30 per thousand images on Haiku 4.5 and $6.48 on Opus 5. Q: Why did my image costs rise after upgrading the model? A: Resolution tiers. Claude 4.7 and later allow up to 4,784 visual tokens per image, earlier models cap at 1,568. The same 4K image is 4,784 tokens on the newer tier against 1,560 on the older one, a factor of 3.07, with no change to your code or your images. Q: Does resizing an image before sending it save money? A: Yes, and proportionally. Cost tracks the patch count, which tracks pixels, so halving both dimensions cuts visual tokens to roughly a quarter. Above the tier limit the provider downscales anyway, so sending an oversized file buys nothing and costs upload time. Q: How do I budget for a vision feature? A: Take your median and worst-case image dimensions, convert both to visual tokens, and price them at your model's input rate against expected volume. The worst case matters more than usual here, because a product accepting uploads has no control over what arrives. ## Sources - Anthropic, vision and image token cost: https://platform.claude.com/docs/en/build-with-claude/vision - Anthropic pricing: https://platform.claude.com/docs/en/docs/about-claude/pricing Run the free Cost Leak Scan: https://app.getculpa.com/scan?source=pseo&slug=image-token-cost&cluster=problem Machine-readable index of every guide: https://getculpa.com/api/pages Human-readable index of every guide: https://getculpa.com/guides Site overview: https://app.getculpa.com/llms.txt Privacy: Culpa runs on your infrastructure. Your prompts and responses never leave it. Culpa counts calls to run your plan, and it fails open, so if it ever breaks your app keeps running.