Image generation API cost calculator
Ballpark your monthly bill for hosted image APIs at fixed per-image rates (DALL·E 3, gpt-image-1, Imagen-style endpoints). Newer token-billed multimodal models work differently — for those, use the text/token calculator instead. Imagen entries are approximate; confirm against Google AI pricing before committing budget.
Preset rates last reviewed: April 2026
Your usage
Square, standard quality
Average across user traffic, marketing batches, or in-app generations.
Estimated cost
Per image (this preset)
$0.040
Estimated per month (~30 days)
$600.00
Per year: $7,200.00
Not included
Moderation, storage (S3, CDN), retries, or token-based image models billed like LLMs. Add those separately.
Image generation is the easiest place for free-tier abuse to hide.
One image looks cheap. A power user generating 200 a day quietly costs more than your highest-paid customer pays. PerUnit attributes every image back to the customer who triggered it, so the cap and pricing decisions are based on who's actually consuming what.
Get early access to PerUnitFrequently asked questions
- How is image generation priced compared to text APIs?
- Most image APIs (DALL·E 3, classic Imagen) are billed flat per image — typically $0.02 to $0.12 per image depending on resolution and quality. Text APIs are billed per token. Newer multimodal models like gpt-image-1 are starting to charge by token instead, which makes them more expensive at scale than the per-image rate suggests. Read the pricing page carefully before assuming a per-image SKU.
- Why are some image models so much more expensive than others?
- Two main drivers: resolution and quality. A 1024×1024 standard-quality image costs roughly $0.02–$0.04; the same image at HD or larger sizes is 2–4× more. Models with stronger visual quality (DALL·E 3 HD, gpt-image-1 high) sit at the top end. The cheapest legitimate option for high-volume work is usually the lowest-resolution standard preset.
- How do I keep image generation costs from running away?
- Three levers, in order of impact: (1) cap free-tier users at a small monthly image quota — image gen is the most-abused free feature in any product that ships it; (2) default to the smallest acceptable resolution and let users opt up if they need more; (3) cache aggressively — if the same prompt produces a usable image, store it and don't regenerate. The first one is usually 10× the impact of the other two combined.
Get a monthly email when AI model pricing changes
One email a month. New model launches, price cuts, deprecations across OpenAI, Anthropic, and Google. No spam, unsubscribe any time.