Looking for text-model pricing? This page compares image models. For per-token prices of chat models (input vs output, billed per million tokens) see the LLM API price comparison, and for the live numbers we charge see the price list. Plain-language answers to "what does API pricing mean" and "how are models priced" live in that comparison page.
Two image APIs can both say "cheap" and mean different things, because image APIs do not share one billing unit. Some bill you per generated image. Others bill you per image token, the same way text models bill per text token. You cannot put a $0.02-per-image figure next to a $30-per-1M-image-token figure and learn which costs less — the units do not line up.
The second thing that skews cheap-image comparisons is quality and resolution tiers. A provider that looks expensive per image may be quoting a high-resolution tier, while a cheaper-looking one quotes a low-resolution draft. You have to compare the same size and quality before the number means anything.
A per-image price is already in the unit you care about: one finished picture, one billable unit. If a model costs $0.04 per image, one generated image is $0.04, two are $0.08, and you can predict 1,000 images cost about $40 before you send anything.
A per-token price is not in that unit yet. You pay for image tokens, and the token count depends on resolution, quality tier, and how the model compresses the output. A 1024×1024 image and a 1536×1024 image use different token counts, so the same model can cost $0.011 for one image and $0.25 for another depending on the tier you request.
That is why a headline like "$30 per 1M output tokens" tells you almost nothing about per-image cost until you know the token cost of the size and quality you need. The comparison only becomes valid after one multiplication: tokens per image, at your target size, times the per-token rate.
Two failure patterns follow. First, comparing a per-image price to a per-token price as if the units matched — two columns pasted side by side without a shared denominator. Second, comparing different tiers. A "cheap" per-image figure often describes the smallest, lowest-quality output, not the size a real product ships.
To get every image API onto one denominator, do three things in order.
The result is a dollars-per-image number you can compare across every provider.
Here is a comparison table built on list prices published by each vendor, converted to a cost-per-generated-image basis where the vendor publishes enough detail. List prices change often and by tier; treat these as a starting point, not a contract.
| Model | Billing unit | List price (as published) | Tapix AI pricing basis |
|---|---|---|---|
| OpenAI GPT Image 2.6 (Sunburst) | Per image token | $8.00 input / $30.00 output per 1M image tokens | List price minus up to 30%, shown beside the list price |
| Google Imagen 4 | Per image | $0.04 per image ($0.02 Fast, $0.06 Ultra) | List price minus up to 30%, shown beside the list price |
| Black Forest Labs FLUX Pro 1.1 | Per image | $0.04 per image | List price minus up to 30%, shown beside the list price |
| xAI grok-2-image | Per image | $0.07 per image | List price minus up to 30%, shown beside the list price |
Two honest notes about this table.
The OpenAI row is the only one billed per token, and that is the whole point: it cannot be placed next to the $0.04-per-image rows without first converting token cost to a per-image cost at a stated size and quality. The per-image rows are already in the right unit.
Google Imagen 4 was retired from the Gemini API on August 17, 2026, so treat its figures as a historical example of the per-image model rather than a current buying option. Check each vendor's pricing page before relying on any number.
One pricing detail matters more for image APIs than for text, because image generation is slower and gets retried more often. A timeout or rate-limit retry on a text model usually costs a few input tokens. On an image model, a failed generation can otherwise waste a full per-image charge — and automatic retries multiply that waste.
Tapix AI bills only for images that are actually generated and returned. A request that fails, times out, or returns an error does not charge your balance, and a retry of that request is only billed when it succeeds. This applies to images the same way it applies to chat.
If you are comparing providers, make this part of the comparison. Ask the documentation: does a failed image request cost anything, and is a retried request billed once or per attempt.
This comparison is most useful to three kinds of teams.
It is not for anyone who needs a signed contract, a data processing agreement, or named support — a discount does not replace those, and we do not offer them.
Why do some image APIs bill per token instead of per image?
Token-based billing lets one endpoint also handle image understanding and editing, so the vendor meters a shared token unit rather than a fixed per-image rate. Per-image billing is simpler to predict but less flexible.
How do I convert a per-token image price to per-image?
Get the token count for your target resolution and quality tier from the provider, then multiply tokens per image by the dollars-per-token rate. The result is a per-image cost you can compare against per-image prices.
Is a lower per-image price always cheaper?
No. A per-image price only matters at the same resolution and quality. A $0.02 figure for a low-quality 1024×1024 image is not cheaper than $0.04 for a high-quality 2K image if your product needs the larger one.
Do I pay for an image that fails to generate?
On Tapix AI, no — failed, timed-out, or errored image requests are not billed, and a retried request is charged only when it succeeds. Check each provider's documentation, because this is not universal.
Which unit should I standardize on when comparing?
Dollars per generated image, at the exact resolution and quality your product ships. Every provider can be converted to this denominator, and it is the only one that lets you compare per-image and per-token APIs side by side.
Image APIs bill by image or by image token, and the two units do not compare until you convert everything to dollars per generated image at a fixed size and quality. Convert first, compare second, and confirm the provider does not bill failed generations. Do that and the "cheap image API" question answers itself with a spreadsheet, not a hunch.
If you want the current per-model numbers, our pricing page shows every image model with its list price beside our price, and the model catalog lists what is available right now.