The short answer: three per-image prices, one shutdown date
OpenAI lists GPT Image 1.5 at $0.009 (low), $0.034 (medium), and $0.133 (high) per 1024×1024 image as of August 24, 2026. Non-square sizes cost more: up to $0.20 for a high-quality 1536×1024 or 1024×1536 output. That is the per-image ladder most pricing pages quote — and it is the least useful number in this article, because GPT Image 1.5 was deprecated on June 2, 2026 and will be removed from the API on December 1, 2026. GPT Image 1.5 pricing, in other words, has a shelf life: any budget you build on the model today is a migration budget in disguise.
So this breakdown has two jobs. First, give you the current official numbers, cleanly dated. Second, give you the pricing decision those numbers actually feed — whether to stay on GPT Image 1.5 through its final months or move to GPT Image 2 now.
We run Modellix, an API aggregator that lists GPT Image 1.5, GPT Image 1.5 Edit, GPT Image 2, and GPT Image 2 Edit routes, and we have a commercial interest when our route appears below. Prices in this article were pulled from OpenAI’s first-party pages and Modellix’s model page on August 24, 2026; where a number could not be verified from a live source, we say so instead of guessing.
Match the price to the thing you are buying
GPT Image 1.5 pricing is quoted two ways, and both are correct:
- Per image — the ladder OpenAI publishes for the three common output sizes (1024×1024, 1024×1536, 1536×1024) at three quality tiers (low, medium, high).
- Per million tokens — the underlying billing rule. The per-image numbers are OpenAI’s precomputed conversion for those common size-and-quality combinations; the token rates are what your account is actually charged, and they matter for anything outside the common cases — especially edits.
Before you use any number in a forecast, fix four things: the exact model ID (gpt-image-1.5 or the snapshot gpt-image-1.5-2025-12-16), the output size, the quality tier, and whether the request generates or edits. An edit bills image input tokens for the source image on top of the output tokens — so “$0.034 per image” is never a universal answer, no matter which page you read it from.
Official GPT Image 1.5 pricing, captured August 24, 2026
The per-image table below is the complete current ladder, read from OpenAI’s GPT Image 1.5 model page and the image generation guide on August 24, 2026.
| Quality | 1024×1024 | 1024×1536 | 1536×1024 |
|---|---|---|---|
| Low | $0.009 | $0.013 | $0.013 |
| Medium | $0.034 | $0.050 | $0.050 |
| High | $0.133 | $0.200 | $0.200 |
Source: OpenAI image-generation guide, accessed August 24, 2026.
The official per-image ladder for GPT Image 1.5 alongside GPT Image 1, GPT Image 1 Mini, and GPT Image 2, captured from OpenAI’s image generation guide on August 24, 2026.
Underneath the ladder, the token rates on the GPT Image 1.5 model page are:
| Component | Rate (per 1M tokens) |
|---|---|
| Text input | $5.00 |
| Cached text input | $1.25 |
| Text output | $10.00 |
| Image input | $8.00 |
| Cached image input | $2.00 |
| Image output | $32.00 |
Source: OpenAI GPT Image 1.5 model page, accessed August 24, 2026.
The per-image figures are consistent with these token rates. A high-quality 1024×1024 output consumes 4,160 image output tokens; 4,160 × $32 / 1,000,000 = $0.133, exactly the published per-image price. You can reproduce any row this way — which is also the right way to estimate a size or workflow OpenAI does not publish a row for.
There is no free image tier for GPT Image 1.5. Every generation is billed; the model’s rate-limit table does not include a Free tier. Azure’s GPT-Image-1.5 Global pricing mirrors the same per-1M-token rates ($5 text input / $1.25 cached / $10 output; $8 image input / $2 cached / $32 image output), which is a useful cross-check if you compare OpenAI-direct against Azure-hosted access.
What GPT Image 2 costs — and what migrating changes
GPT Image 1.5, GPT Image 1 Mini, and chatgpt-image-latest are all scheduled for removal from the API on December 1, 2026; OpenAI’s deprecation schedule names GPT Image 2 as the replacement for all three. GPT Image 1.5 pricing is therefore a question with a shelf life — the more useful question is what the same workload costs on GPT Image 2.
| Model | Quality | 1024×1024 | 1024×1536 | 1536×1024 |
|---|---|---|---|---|
| GPT Image 1.5 | Low | $0.009 | $0.013 | $0.013 |
| Medium | $0.034 | $0.050 | $0.050 | |
| High | $0.133 | $0.200 | $0.200 | |
| GPT Image 2 | Low | $0.006 | $0.005 | $0.005 |
| Medium | $0.053 | $0.041 | $0.041 | |
| High | $0.211 | $0.165 | $0.165 | |
| GPT Image 1 Mini | Low | $0.005 | $0.006 | $0.006 |
| Medium | $0.011 | $0.015 | $0.015 | |
| High | $0.036 | $0.052 | $0.052 |
Source: OpenAI image-generation guide, accessed August 24, 2026.
The direction of the price change depends on the size you request:
- Square (1024×1024), medium and high: GPT Image 2 is more expensive — $0.053 vs $0.034 at medium, and $0.211 vs $0.133 at high.
- Non-square sizes and low quality: GPT Image 2 is cheaper — $0.041 vs $0.050 at medium 1024×1536, $0.165 vs $0.200 at high, and $0.006 vs $0.009 at low.
So migrating is not automatically a budget increase. If your workload is dominated by landscape or portrait outputs, or by low-quality drafts, GPT Image 2 lowers your per-image cost. If it is dominated by square medium or high outputs, the same volume costs more — and that delta is the number to put in front of whoever owns the budget before December 1.
Two non-price differences belong in the same decision. Transparent backgrounds are available in preview for GPT Image 2 — set background: "transparent" and output png (the default) or webp; jpeg is not supported with transparent backgrounds. Support is model-dependent, so confirm the current capability on the live guide before you build on it. And GPT Image 2 meters output through a newer token calculator rather than the fixed legacy counts, so for unusual sizes use OpenAI’s calculator rather than the legacy arithmetic. For capabilities, rate limits, and working code for both models, our GPT Image 1.5 API guide and GPT Image 2 API guide go deeper.
Direct API or an aggregator: compare routes, not brand names
GPT Image 1.5 pricing also depends on which route you call the model through. OpenAI sells direct access; several aggregators resell the same model. A route comparison only means something when the model ID, size, quality, and billing unit match — comparing a per-image ladder to a per-call fee to a credits balance is comparing different products.
| Route | Current displayed price (Aug 24, 2026) | What you get | Watch out for |
|---|---|---|---|
| OpenAI direct | $0.009–$0.200 per image (official ladder) | First-party model access, OpenAI billing | Organization verification may be required to enable GPT Image models |
| fal.ai | $0.009–$0.200 per image (official ladder) | Serverless inference, no OpenAI org setup | Same underlying model; adds its own text-token fees ($0.005/1k input, $0.010/1k output) on top |
| Together AI | $0.034 per 1024×1024 medium image | Hosted access with an OpenAI-compatible surface | Confirm the current list on Together’s model page |
| Modellix | $0.0138–$0.2334 per image (displayed range) | One key across providers, per-call cost logs, no China-account needed for Chinese models | Distribution layer, not a model owner; range spans sizes and qualities |
Sources: OpenAI model page and image-generation guide (linked above), the fal.ai model page, the Together AI model page, and the Modellix GPT Image 1.5 model page — all accessed August 24, 2026.
The honest boundary: Modellix does not operate its own image model — it passes through providers’ economics and charges per call, so the range on its GPT Image 1.5 model page sits around the official ladder, not below it. The displayed low end ($0.0138) is above OpenAI’s $0.009 low-quality rate, so this is not a “cheaper than OpenAI” claim; what a shared key buys is one integration while you evaluate, plus per-call cost logs that let you compare routes on real usage. If you need OpenAI’s organization verification, a single-provider SLA, or one vendor to hold accountable, direct is the simpler answer. For the broader workflow of routing many models through one integration, see our unified AI API overview and the OpenAI provider page for the current catalog.
Route comparison — illustrative Modellix artwork, not a product screenshot. It shows the decision shape, not a price list.
Budget math for a real workload
Using the official per-image figures (output only; prompt text tokens excluded), a square medium-quality workload looks like this:
| Volume | 1024×1024 medium | 1024×1024 high |
|---|---|---|
| 1,000 images | $34 | $133 |
| 10,000 images | $340 | $1,330 |
Derived from OpenAI’s official per-image rates, accessed August 24, 2026. Add text input tokens and, for edits, image input tokens.
For the same workload after migration to GPT Image 2, square medium goes to $53 per 1,000 images and $530 per 10,000 — the budget conversation most teams need to have before December 1. Edits are the other budget trap: every edit bills image input tokens for the source image at $8 per 1M, plus output tokens, so a heavy edit pipeline can cost several times an equivalent generate-only flow. A concrete example, using OpenAI’s token rates and the guide’s token counts (medium 1024×1024 = 1,056 image output tokens):
| Task | Output cost | Image input cost | Total |
|---|---|---|---|
| 1,000 generated images (medium, 1024×1024) | $34 (official per-image) | — | $34 |
| 1,000 edits (medium, 1024×1024, ~1,000 input image tokens each) | 1,056 × $32 / 1M × 1,000 = $34 | 1,000 × $8 / 1M × 1,000 = $8 | ~$42 |
Worked from OpenAI’s token rates (image output $32 per 1M, image input $8 per 1M) and the guide’s medium 1024×1024 output-token count, accessed August 24, 2026. Real source-image token counts vary with resolution and fidelity.
OpenAI’s pricing page currently lists batch rates for GPT Image 2 (image input $4 / cached $1 / output $15 per 1M tokens) and no longer lists GPT Image 1.5 rows; check the live pricing page before assuming a batch discount for legacy models.
A practical checklist
- Name the model and snapshot.
gpt-image-1.5resolves togpt-image-1.5-2025-12-16; pin the snapshot if you need stable behavior. - Record the billing unit. Keep per-image rows and per-1M-token rates in separate columns of your budget.
- Match size and quality before comparing. Square medium on GPT Image 1.5 costs $0.034; the same quality at 1024×1536 costs $0.050.
- Count edits as input plus output. An edit bills source-image tokens plus output tokens.
- Set the migration window. December 1, 2026 is the removal date; plan the flip while GPT Image 1.5 is still callable.
- Compare routes only on matched specs. Then add the operational value of the route — verification, billing, logs.
- Recheck the source before purchase. Pricing and availability pages change; this article is a dated breakdown, not a quote.
Check GPT Image 1.5 and GPT Image 2 Route Pricing
Log in to compare current GPT Image route prices and per-call cost logs on one Modellix key.
LoginFAQ
How much does GPT Image 1.5 cost per image?
As of August 24, 2026, OpenAI lists $0.009 (low), $0.034 (medium), and $0.133 (high) for 1024×1024 outputs, and $0.013 / $0.050 / $0.200 for the same tiers at 1024×1536 and 1536×1024. There is no free image tier.
Why does one page say $0.034 per image and another says $5 per million tokens?
Both are official. The per-image figures are OpenAI’s conversion for common size-and-quality combinations; the token rates ($5 text input, $10 text output, $8 image input, $32 image output per 1M tokens) are what the account is actually billed. The per-image number is only universal for the exact size and quality it describes.
Is GPT Image 1.5 editing more expensive than generation?
Usually, because an edit request also bills image input tokens for the source image ($8 per 1M) on top of the output tokens. Budget edits as input-plus-output rather than at the per-image generate rate.
Is there a free way to use GPT Image 1.5?
No. Every generation is billed; the model’s rate-limit table has no Free tier. “GPT Image 1.5 free” queries usually surface ChatGPT subscriptions, which are a different product from API access.
Is GPT Image 2 free?
No — same story as GPT Image 1.5. GPT Image 2’s rate-limit table lists Free as “Not supported”, so every generation is billed; OpenAI’s pricing page is the current reference for rates.
When was GPT Image 1.5 released?
The API snapshot is gpt-image-1.5-2025-12-16, which pins the model’s lineage to December 16, 2025 — the same snapshot gpt-image-1.5 resolves to today.
When does GPT Image 1.5 shut down?
It was deprecated on June 2, 2026 and is scheduled for removal from the API on December 1, 2026, together with GPT Image 1 Mini and chatgpt-image-latest. OpenAI’s deprecation schedule lists GPT Image 2 as the replacement.
Is DALL·E 3 still an option?
No. OpenAI removed the DALL·E snapshot models (dall-e-2 and dall-e-3) from the API on May 12, 2026, directing users to the GPT Image family — so DALL·E 3 is not part of any current pricing comparison.
How does GPT Image 1.5 compare to Nano Banana Pro?
That is a capability question, not a pricing one, so this article deliberately does not expand it. Nano Banana Pro is Google’s Gemini image model — compare capabilities on the official model pages; the numbers above only price GPT Image 1.5 and GPT Image 2 API workloads.
Is Modellix cheaper than OpenAI direct for GPT Image 1.5?
Not as a blanket statement. Modellix’s displayed range is $0.0138–$0.2334 per image, with a low end above OpenAI’s $0.009 low-quality rate. It is a distribution layer with per-call cost logging — useful when you want one key and one bill across providers, not a guaranteed lower unit price.
Pricing sources accessed August 24, 2026: OpenAI GPT Image 1.5 model page, OpenAI pricing page, OpenAI image-generation guide, OpenAI deprecation schedule, and the Modellix GPT Image 1.5 model page. OpenAI and Modellix can change pricing, availability, and terms without notice. Modellix is an API aggregator with a commercial interest in this comparison and does not operate its own image models. Access image and video models, including leading Chinese models, through a single API key at modellix.ai.
