Danke,
Most of my images are not art, but are for entertainment purposes only and should not be taken seriously.
1 Like
Respect - I’m not a police man. 
Selbst habe ich manchmal Gedanken, was mit der Zukunft noch kommt.
Zum Beispiel in Europa.
Are you from the DACH region?
For many countries is the self language LLM is a problem.
And the rules are others, but medically would I say too: openai first…
1 Like
Respect - I’m not a police man.
I know. I was just making a joke.
Are you from the DACH region?
No. I learned a bit of German in high school many years ago…
I believe that your profession, and heath care in general, will benefit greatly from AI in the not too distant future - lets hope!
Have a Great Holiday!
2 Likes
OpenAI still has not touched on any documentation of costs for gpt-image-1.5, which additionally includes:
-
reasoning text (undelivered) output tokens (200-500 typical)
-
mandatory or non-working input_fidelity switch, always “high”
Documentation implies the switch should work - a) GPT Image models (gpt-image-1.5, gpt-image-1, and gpt-image-1-mini) support high input fidelity. b) If you are using gpt-image-1.5, the first 5 input images will be preserved with higher fidelity. c) To enable high input fidelity, set the input_fidelity parameter to high. The default value is low.
-
n>1: you pay again for input for each generation, unlike dall-e or chat
edits: n=1 "input_tokens_details": {"image_tokens": 4354,"text_tokens": 59} edits: n=2 "input_tokens_details": {"image_tokens": 8708,"text_tokens": 118} -
There is no usage report of cached, nor API shape for it in yaml spec, yet cached input is in the price list. Despite extensive repeated API calls in app development, I’ve not once received a single cached input token, with assurance I’m looking in the right place:
- beyond denials: costs of dropped images from safety inspection - I’ve gotten just one of two requested, for example.
Consolidated pricing - calculated
gpt-image-1.5 per-image input pricing by size (n=1)
| Aspect ratio band (AR = long/short) | Resized short×long threshold range | Tiles | Token formula | Total tokens | Cost per image |
|---|---|---|---|---|---|
| AR = 1.00 (exact square) | 512×512 | 1 | 65 + 129×1 + 4160 | 4354 | $0.034832 |
| 1.00 < AR ≤ 1.25 (square-ish) | 512×(512–640] | 2 | 65 + 129×2 + 4160 | 4483 | $0.035864 |
| 1.25 < AR ≤ 2.00 (rectangular) | 512×(640–1024] | 2 | 65 + 129×2 + 6240 | 6563 | $0.052504 |
| 2.00 < AR ≤ 3.00 (more rectangular) | 512×(1024–1536] | 3 | 65 + 129×3 + 6240 | 6692 | $0.053536 |
| 3.00 < AR ≤ 4.00 (very rectangular) | 512×(1536–2048] | 4 | 65 + 129×4 + 6240 | 6821 | $0.054568 |
-
this assumes shorter dimension >=512px (OpenAI’s downsize), and that ‘closer to square’ is actually true in regards to image fidelity
-
16 image inputs are allowed on API; assume the max that can be scraped from chat by the Responses tool is similar.
Per-image output pricing by size (n=1)
| Model | Quality | 1024×1024 | 1024×1536 | 1536×1024 |
|---|---|---|---|---|
| GPT Image 1.5 | Low | $0.008704 | $0.013056 | $0.012800 |
| GPT Image 1.5 | Medium | $0.033792 | $0.050688 | $0.050176 |
| GPT Image 1.5 | High | $0.133120 | $0.199680 | $0.198656 |
| GPT Image 1 | Low | $0.010880 | $0.016320 | $0.016000 |
| GPT Image 1 | Medium | $0.042240 | $0.063360 | $0.062720 |
| GPT Image 1 | High | $0.166400 | $0.249600 | $0.248320 |
| GPT Image 1 Mini | Low | $0.002176 | $0.003264 | $0.003200 |
| GPT Image 1 Mini | Medium | $0.008448 | $0.012672 | $0.012544 |
| GPT Image 1 Mini | High | $0.033280 | $0.049920 | $0.049664 |
Fixed-price image models
DALL·E 3
| Quality | 1024×1024 | 1024×1792 | 1792×1024 |
|---|---|---|---|
| Standard | $0.04 | $0.08 | $0.08 |
| HD | $0.08 | $0.12 | $0.12 |
DALL·E 2
| Quality | 256×256 | 512×512 | 1024×1024 |
|---|---|---|---|
| Standard | $0.016 | $0.018 | $0.02 |
Distilled token pricing sheet you are offered
Text token pricing (per 1M tokens)
| Model | Input | Cached input | Output |
|---|---|---|---|
| gpt-image-1.5 | $5.00 | $1.25 | $10.00 |
| gpt-image-1 | $5.00 | $1.25 | — |
| gpt-image-1-mini | $2.00 | $0.20 | — |
Image token pricing (per 1M tokens)
| Model | Input | Cached input | Output |
|---|---|---|---|
| gpt-image-1.5 | $8.00 | $2.00 | $32.00 |
| gpt-image-1 | $10.00 | $2.50 | $40.00 |
| gpt-image-1-mini | $2.50 | $0.25 | $8.00 |
Example cost-avoidance “edit”
gpt-image-1.5, 1 square + 3 rectangular in, quality:medium 1024x1024 out
- Text input: 122 × 0.000005 = $0.000610000000
- Image input: 24043 × 0.000008 = $0.192344000000
- Text output: 448 × 0.000010 = $0.004480000000
- Image output: 1056 × 0.000032 = $0.033792000000
Total cost: $0.231226
4 Likes
Another bump - of October’s report:
Bumping this in 2026, because the docs are still wrong about the per-image price grid for the mini model still, where nothing seems to account for this except for bad math (or hidden fees).
1: OpenAI’s output token consumption for “the models”, quality vs resolution
| Quality | Square (1024×1024) | Portrait (1024×1536) | Landscape (1536×1024) |
|---|---|---|---|
| Low | 272 tokens | 408 tokens | 400 tokens |
| Medium | 1056 tokens | 1584 tokens | 1568 tokens |
| High | 4160 tokens | 6240 tokens | 6208 tokens |
1a: My token results by running a generate at each, from “usage”
- identical
| Quality | Square (1024×1024) | Portrait (1024×1536) | Landscape (1536×1024) |
|---|---|---|---|
| Low | 272 tokens | 408 tokens | 400 tokens |
| Medium | 1056 tokens | 1584 tokens | 1568 tokens |
| High | 4160 tokens | 6240 tokens | 6208 tokens |
2: OpenAI per-image pricing page for gpt-image-1-mini (output)
| Quality | 1024×1024 | 1024×1536 | 1536×1024 |
|---|---|---|---|
| Low | $0.005 | $0.006 | $0.006 |
| Medium | $0.011 | $0.015 | $0.015 |
| High | $0.036 | $0.052 | $0.052 |
3: Calculated per-image pricing for gpt-image-1-mini using Table #1 tokens and $8.00 / 1M output image tokens
| Quality | 1024×1024 | 1024×1536 | 1536×1024 |
|---|---|---|---|
| Low | $0.002176 | $0.003264 | $0.003200 |
| Medium | $0.008448 | $0.012672 | $0.012544 |
| High | $0.033280 | $0.049920 | $0.049664 |
OpenAI’s pricing page is thus showing us prices vs actual:
| Quality | 1024×1024 | 1024×1536 | 1536×1024 |
|---|---|---|---|
| Low | 229.8% | 183.8% | 187.5% |
| Medium | 130.2% | 118.4% | 119.6% |
| High | 108.2% | 104.2% | 104.7% |
Some pics
gpt-image-1-mini @ 1536 and quality:high - about $0.05
dall-e-3 @ 1792 and not HD - $0.08
The second failure in a day to outfill the sides (the first completely white beyond 1024 width)
A single project request for 100 images might show us what’s being billed in dollars with the needed accuracy…
1 Like
Wow, really good analysis. Good job!
1 Like