Searching “openai image generation api pricing” in August 2026 lands you in a weird place: OpenAI’s official pricing page gives you tokens per million, its image generation guide hides the per-image table one click deeper, and the rest of the top results are 2025 forum threads quoting prices for models that no longer exist. The number you actually came for — what one image costs — is never on the same page as the models you can call today.
This article fixes that with one dated snapshot. Every price below was pulled from first-party sources on August 25, 2026: OpenAI’s API pricing page, its image generation guide, and its model deprecations page. You also get worked per-image costs, the models that are about to disappear, and an honest look at direct versus aggregator routes. Modellix is an AI model API aggregator and has a commercial interest in this comparison — we carry OpenAI image routes on our platform, so the aggregator section compares billing units honestly instead of declaring a winner.
What OpenAI image generation API pricing actually is in 2026
OpenAI’s image models are priced per token, not per image. A request bills three token types: text input tokens (your prompt), image input tokens (reference images on edit calls), and image output tokens (the generated image). Each model has its own rate per million tokens, and the final cost of one image is the number of output tokens it consumes times that rate.
The current lineup, from the official pricing page (standard tier, accessed August 25, 2026):
| Model | Image input | Image output | Text input | Status |
|---|---|---|---|---|
| gpt-image-2 | $8.00 | $30.00 | $5.00 | Current flagship |
| gpt-image-1.5 | $8.00 | $32.00 | $5.00 | Previous generation |
| gpt-image-1-mini | $2.50 | $8.00 | $2.00 | Cost-efficient, deprecated |
| gpt-image-1 | $10.00 | $40.00 | $5.00 | Previous generation |
| chatgpt-image-latest | $8.00 | $32.00 | $5.00 | ChatGPT surface model, deprecated |
Source: OpenAI API pricing, accessed August 25, 2026 — the page currently renders gpt-image-2; the legacy model rates above were verified individually on their model detail pages the same day. Cached-input rates exist but are omitted for brevity; batch API pricing is 50% of these standard rates.
Two things worth noting before you compare numbers anywhere. First, OpenAI’s own models page describes gpt-image-2 as the “state-of-the-art image generation model” — everything else in the table above is a legacy or transitional listing. Second, the model you call matters more than the rate card: gpt-image-1-mini’s output tokens cost one-fifth of gpt-image-1’s, but its quality ceiling is lower, and it is scheduled for removal (more below). For context on how the whole family fits together, our OpenAI image generation API guide covers the endpoints and parameters.
What one image really costs: quality, size, and token math
Because output tokens scale with complexity, OpenAI publishes token counts per quality setting and size. For the token-based family (gpt-image-1 and gpt-image-1.5), a 1024×1024 image consumes roughly 272 output tokens at low quality, 1,056 at medium, and 4,160 at high (gpt-image-1-mini uses its own per-image rate card instead of these token counts). Multiply by the model’s output rate and you get the official per-image figures — for example, gpt-image-1 at high quality: 4,160 tokens × $40/M ≈ $0.167, exactly what OpenAI’s per-image table lists.
The official guide’s per-image table (accessed August 25, 2026) is the number to budget with. Square 1024×1024 shown; non-square sizes shift the cost slightly (a larger 1536×1024 image can sometimes cost fewer output tokens than square at the same quality):
| Model | Low (1024×1024) | Medium (1024×1024) | High (1024×1024) |
|---|---|---|---|
| gpt-image-2 | $0.006 | $0.053 | $0.211 |
| gpt-image-1.5 | $0.009 | $0.034 | $0.133 |
| gpt-image-1 | $0.011 | $0.042 | $0.167 |
| gpt-image-1-mini | $0.005 | $0.011 | $0.036 |
Source: OpenAI image generation guide, accessed August 25, 2026. Figures are per generated image at the listed quality and size; text and image input tokens still add to the request total.
How a request is billed: text and image input tokens enter the request, the model renders at a quality tier, and output tokens are charged against the model’s per-million rate. Illustrative Modellix concept graphic; no prices shown because rates change faster than images.
The per-image figures are the reason a single “OpenAI image price” answer is wrong: the same model spans $0.006 to $0.211 per image depending on quality, a 35× spread. If you are budgeting an integration, decide the quality floor first, then price the tier — not the model name.
Which models are still live — and which are going away
The deprecation schedule is the most under-covered part of this topic, and it changes which model you should build on. From OpenAI’s deprecations page, accessed August 25, 2026:
| Model | Removal / replacement date | Replacement |
|---|---|---|
| dall-e-2, dall-e-3 | Removed May 12, 2026 | gpt-image-2 / gpt-image-1 / gpt-image-1-mini |
| gpt-image-1 | October 23, 2026 | gpt-image-2 |
| gpt-image-1-mini | December 1, 2026 | gpt-image-2 |
| gpt-image-1.5 | December 1, 2026 | gpt-image-2 |
| chatgpt-image-latest | December 1, 2026 | gpt-image-2 |
Source: OpenAI model deprecations, accessed August 25, 2026.
Three practical takeaways. DALL·E is dead: any tutorial or 2025 blog post that hands you model: "dall-e-3" will fail today — that is why “dall-e api pricing” searches now need to land on gpt-image-2 answers. gpt-image-1 and the mini/1.5 variants still work but have fixed removal dates inside 2026, so a new integration should default to gpt-image-2. And the historical context matters for pricing: when OpenAI launched gpt-image-1 in April 2025, its announcement described per-token pricing at $5 text input / $10 image input / $40 image output per million tokens, “roughly $0.02, $0.07, and $0.19 per generated image” for low, medium, and high square images — the shift from DALL·E’s flat per-image pricing to token pricing is why today’s per-image cost varies with quality at all. Our separate gpt image 1.5 pricing post has the mid-generation numbers if you are comparing legacy choices.
What 1,000 images actually cost: worked examples
Concrete numbers beat abstract rates. At gpt-image-2’s official per-image equivalents (1024×1024), 1,000 images cost:
| Quality | 1,000 images, standard | 1,000 images, batch |
|---|---|---|
| Low | $6 | $3 |
| Medium | $53 | $26.50 |
| High | $211 | $105.50 |
Derived from the official per-image table (accessed August 25, 2026); batch pricing is 50% of standard. Input tokens for prompts add a negligible amount (see below).
The batch row is easy to miss and worth building around: OpenAI’s pricing page lists batch API usage at half the standard rate for image models. If your workload tolerates 24-hour completion windows, asynchronous batch generation halves your image line item outright.
Edit requests cost more than fresh generations, and this is where budgets blow up. gpt-image-2 processes every input image at high fidelity automatically, so an edit call with reference images can carry substantially more image input tokens than a text-only generation. Each streamed partial image also adds 100 output tokens. The rule: for edit-heavy workloads, model the input-image tokens into your unit cost — the per-image table alone understates the total.
For the token-math question behind most budget spreadsheets: at gpt-image-2 standard rates, 1,000 tokens cost $0.005 as text input, $0.008 as image input, and $0.03 as image output. For the token-based family (gpt-image-1, gpt-image-1.5), OpenAI publishes the output-token counts — roughly 272 tokens for a low-quality 1024×1024 image, 1,056 for medium, and 4,160 for high; multiply by the model’s output rate and you reproduce the per-image table. gpt-image-2 is priced from its published per-image figures directly, which is why the official guide tells you to use the calculator rather than the older token estimates for it.
Direct OpenAI vs an aggregator: compare routes, not brands
Aggregators resell the same models behind their own billing. On August 25, 2026, Modellix’s OpenAI provider page displayed these per-image route ranges:
| Modellix route | Displayed range (per image) |
|---|---|
| openai/gpt-image-2 | $0.0041–$0.3943 |
| openai/gpt-image-2-edit | $0.0117–$0.4019 |
| openai/gpt-image-1.5 | $0.0138–$0.2334 |
| openai/gpt-image-1.5-edit | $0.0138–$0.2334 |
Source: Modellix model pages, accessed August 25, 2026. Ranges cover the full size/quality matrix, including sizes beyond OpenAI’s three standard comparison dimensions; the low end reflects low-quality smaller outputs, the high end reflects high-quality larger outputs.
The two routes in one picture: a direct line from one vendor to your application versus one key fanning out through a hub to many providers. Illustrative Modellix concept graphic; no prices shown.
Read those ranges honestly. They are not a claim that the aggregator is cheaper: the low end of a Modellix range can sit below OpenAI’s standard-tier equivalents because the range spans more sizes and qualities, and the high end can sit above them for the same reason. On matched official rates, direct OpenAI is usually the floor — an aggregator cannot beat the vendor’s own price for the same token consumption, because it pays the vendor. What a route actually buys you is workflow: one API key and one bill for models from many providers, per-call cost logs instead of monthly invoices per vendor, and access to Chinese image and video models without a China-region account. OpenRouter, the other major aggregator, prices the same OpenAI image model at the official $8 input / $30 image output per million tokens on its GPT-5.4 Image 2 page — the aggregator’s margin lives in operational convenience and routing, not in undercutting the vendor.
For a real comparison, match the exact model, quality, size, and billing unit on both sides, then decide whether the integration path is worth the difference. Modellix’s pricing page and model catalog are the current reference for what we display; we do not set or discount OpenAI’s official rates.
Start Generating with OpenAI on Modellix
Log in to call gpt-image-2 and 30+ other image and video models with one API key.
LoginWhen to go direct to OpenAI — and when the aggregator route wins
Go direct to OpenAI when any of these is true: you are committing to OpenAI as your sole image vendor and want the vendor’s own account, throughput, and enterprise terms; you need data-residency regions or enterprise contracts that aggregators do not resell; you want exact token-level accounting on every request; or you need launch-day access to a brand-new model the moment it ships.
Use an aggregator route when you are evaluating models across vendors — comparing gpt-image-2 against Google’s or Chinese providers’ image models with the same prompt — or when you want one billing relationship, one key, per-call cost logs, and no per-vendor signup. For cost modeling across the wider market, our cheapest AI API comparison shows how per-call pricing varies between providers for the same job.
Neither path is universally cheaper; that is the point. Direct gives you the vendor floor and full control; aggregators give you optionality and consolidated operations. Price the exact workload on both, then choose on fit, not on brand loyalty to either side.
A practical checklist before you commit
- Name the model and route. “OpenAI image generation” is not a price; gpt-image-2, gpt-image-1.5, and gpt-image-1-mini have different rates and different removal dates.
- Decide quality and size first. The same model spans $0.006–$0.211 per image at 1024×1024; your quality floor defines your unit cost.
- Count input tokens on edits. gpt-image-2 processes reference images at high fidelity; edit workloads add image input tokens on top of output.
- Use batch for non-urgent jobs. The batch API halves the standard image rate when a 24-hour window is acceptable.
- Log per-call cost from day one. Aggregators expose per-task cost logs; direct accounts need your own metering — either way, budget on logged averages, not the rate card.
- Recheck before purchase. Pricing and deprecation pages change; this article is a dated snapshot, not a quote.
FAQ
Is OpenAI image generation free?
No. The API is strictly pay-as-you-go — there is no free image generation tier, and every request bills text input, image input, and image output tokens. New-account credits, when offered, are small and consumed quickly by image output tokens; check your platform billing page for your actual balance.
How much do 1,000 tokens cost?
At gpt-image-2 standard rates (August 25, 2026): 1,000 text input tokens cost $0.005, 1,000 image input tokens cost $0.008, and 1,000 image output tokens cost $0.03. For scale, the token-based family (gpt-image-1 and gpt-image-1.5) uses roughly 272 output tokens for a low-quality 1024×1024 image and about 4,160 for a high-quality one; for gpt-image-2 itself, budget from its official per-image figures instead of the older token estimates.
Why did OpenAI move from per-image to per-token pricing?
DALL·E models billed a flat price per image by resolution. With gpt-image-1 (April 2025), OpenAI switched to token-based pricing so that cost scales with the actual complexity of the output — more tokens for bigger, higher-quality images, fewer for drafts. That is why “one image” no longer has a single price.
What is the cheapest OpenAI image model?
By token rate, gpt-image-1-mini: $2.50 image input / $8 image output per million tokens, or roughly $0.005–$0.036 per 1024×1024 image depending on quality. It is scheduled for removal on December 1, 2026, so new builds should treat gpt-image-2 at low quality ($0.006/image) as the durable budget option.
Does image editing cost more than plain generation?
Usually, yes. Edit calls add image input tokens for reference images, and gpt-image-2 processes every image input at high fidelity, which raises input token counts. Streamed partial images add 100 output tokens each. Model input tokens into edit-heavy budgets.
Is gpt-image-2 more expensive than gpt-image-1?
It depends on quality. At high quality, gpt-image-2 is more expensive ($0.211 vs $0.167 per 1024×1024 image); at low quality it is cheaper ($0.006 vs $0.011). The models also differ in capability — gpt-image-2 is the current flagship and the replacement for every older model.
Can I get OpenAI image generation cheaper through an aggregator?
Not as a blanket claim. On matched model, quality, and size, direct OpenAI pricing is the floor; aggregator ranges span more sizes and can look lower or higher at the extremes. Aggregators compete on workflow — one key, one bill, multi-vendor access — not on undercutting the vendor.
Pricing sources accessed August 25, 2026: OpenAI API pricing, OpenAI image generation guide, OpenAI model deprecations, and the Modellix OpenAI provider page. OpenAI and Modellix can change pricing, availability, and terms without notice; validate against each provider’s live pricing before committing. Access image and video models, including the leading Chinese models, through a single API key at modellix.ai.
Cover image: illustrative Modellix artwork; it is not an OpenAI product screenshot or source evidence.
