GPT Image 2 (OpenAI's gpt-image-2) costs between about half a cent and about $0.55 per image depending on two things: the quality setting and the host. At high quality, the default nearly everywhere, a square image is $0.211 on fal at 1024 x 1024, $0.128 on Replicate (one price for every size), 20 Runway API credits ($0.20) at 1K or 2K, and on 8frame 29 credits, about $0.55 on the $19 Starter plan and $0.47 on the $49 Creator plan. At low quality the same image is $0.006 on fal, $0.012 on Replicate, 1 Runway credit and 2 credits on 8frame. OpenAI itself publishes GPT Image 2 only as token prices. Every number is from the vendor's own page, checked 2026-10-21.
Quality moves the price more than the host
GPT Image 2 has three quality levels, and the gap between them is far larger than the gap between hosts. On fal a 1024 x 1024 image is $0.006 at low, $0.053 at medium and $0.211 at high: high costs about 35 times low. Replicate's three tiers are $0.012, $0.047 and $0.128; Runway's API charges 1, 5 and 20 credits at 1K or 2K. On 8frame the square image is 2, 8 or 29 credits. Before you compare hosts, decide whether the job needs high quality at all; for layout drafts and quick concepts, medium or low changes the bill by an order of magnitude.
Size matters on some hosts and not on others. fal's table and Runway's API both price by size: fal charges $0.145 at high quality for 1024 x 768 and $0.401 for 3840 x 2160; Runway charges 20 credits at 1K or 2K and 41 at 4K, and its "auto" resolution "is billed at the 4K tier". Replicate prints one price per quality level with no size tier. On 8frame the square ratio is the dearest: 29 credits at high, against 20 for 16:9, 9:16, 4:3 and 3:4.
Tokens, not images, on OpenAI. OpenAI's pricing page lists gpt-image-2 at $8.00 per million image input tokens and $30.00 per million image output tokens ($4.00 and $15.00 in Batch), and points to a calculator in its image generation guide for per-image estimates; it prints no per-image figure. fal lists the same token rates and converts them into its per-image table, and rounds every request up "to the closest hundredth of a cent".
The price table
| Host | Vendor's price, as written | Low, square | Medium, square | High, square | What else the page says |
|---|---|---|---|---|---|
OpenAI API (gpt-image-2) |
$8.00 / $30.00 per 1M image input / output tokens; text input $5.00 | Not printed per image | Not printed per image | Not printed per image | Batch $4.00 / $15.00; cached image input $2.00 |
fal (openai/gpt-image-2) |
Per-image table by size and quality | $0.006 (1024 x 1024) | $0.053 | $0.211 | 3840 x 2160: $0.012 / $0.101 / $0.401; quality default "high" |
Replicate (openai/gpt-image-2) |
"per output image": low $0.012, medium $0.047, high $0.128 | $0.012 | $0.047 | $0.128 | "auto" quality billed $0.128; you can bring your own OpenAI key and pay OpenAI directly |
Runway API (gpt_image_2) |
1 / 5 / 20 credits at 1K or 2K, 2 / 11 / 41 at 4K; "$0.01 per credit" | $0.01 | $0.05 | $0.20 | "Default quality is high"; billed per image times the output count |
| 8frame | 2 / 8 / 29 credits for a square image; 2 / 6 / 20 for 16:9, 9:16, 4:3, 3:4 | $0.04 Starter, $0.03 Creator, $0.02 top-up | $0.15, $0.13, $0.08 | $0.55, $0.47, $0.29 | Up to 4 images a run, priced per image; 1 to 10 reference images |
We have not compared outputs across hosts; this page compares prices and published options only.
How the dollar column was computed
- fal, Replicate: no arithmetic, the per-image price is on the page.
- Runway API: credits x $0.01: 20 credits = $0.20, 41 = $0.41.
- 8frame: $19 / 1,000 credits = $0.019 per credit; x 29 = $0.55, x 8 = $0.15, x 2 = $0.04. Creator: $49 / 3,000 = $0.0163; x 29 = $0.47. Top-up packs are $0.01 a credit: $0.29, $0.08, $0.02.
Where 8frame sits
8frame is the dearest host in the table at high quality: 29 credits for a square image is $0.29 even at the top-up rate, against $0.128 on Replicate and $0.20 on Runway's API. Two choices bring it down. The first is the ratio: a 16:9 or 9:16 high-quality image is 20 credits, about $0.38 on Starter. The second is quality: medium is 8 credits square and 6 for the other ratios, low is 2 credits everywhere.
The GPT Image node on 8frame opens on GPT Image 2 at high quality, square (1:1) for a text prompt. A square image comes back at 1024 x 1024; there is no 2K or 4K setting on the node, so the 4K tiers fal and Runway sell are not available on 8frame. Each run makes 1 to 4 images, each billed at the per-image price, and up to 10 reference images do not change it. GPT Image 1.5 stays selectable in the same node. A failed generation is refunded automatically (do failed AI generations cost credits).
The reason to run it on 8frame is the board, not the rate: a GPT Image 2 text-heavy draft next to Nano Banana 2 at 11 credits or Seedream 5 Pro at 7, paid from one balance. If the job is lettering, the side-by-side in Ideogram 4.5 vs GPT Image 2 vs Nano Banana Pro for text is a better starting point than this table; every image model's 8frame price is in AI image generator price per image.
Which host for which job
- Volume through an API at high quality: Replicate, $0.128 per image at any size it offers.
- Large outputs (2K to 4K): fal or Runway's API, which price them explicitly ($0.222 at 2560 x 1440 and $0.401 at 3840 x 2160 on fal at high; 41 Runway credits at 4K).
- Cheap drafts: low quality anywhere: $0.006 on fal, 1 Runway credit, 2 credits on 8frame.
- One balance across many models: 8frame, where the same credits pay for Nano Banana, Seedream, FLUX and the video models.
FAQ
How much does GPT Image 2 cost per image?
At high quality, the default: $0.211 on fal for 1024 x 1024, $0.128 on Replicate, 20 Runway API credits ($0.20) at 1K or 2K, and 29 credits on 8frame for a square image (about $0.55 on Starter, $0.29 at the top-up rate). At low quality it is $0.006 on fal and $0.012 on Replicate.
Where is GPT Image 2 cheapest?
At high quality, Replicate at $0.128 per image. At medium, Replicate ($0.047), Runway's API ($0.05) and fal ($0.053) are within a cent of each other. At low, fal's $0.006 for a square image is the lowest figure printed. OpenAI's own API bills tokens, so its per-image cost depends on the image and is not printed on the pricing page.
Does image size change the GPT Image 2 price?
On fal and Runway's API, yes: fal's high-quality price runs from $0.145 at 1024 x 768 to $0.401 at 3840 x 2160, and Runway charges 41 credits at 4K against 20 at 1K or 2K. Replicate prints one price per quality level. On 8frame the square ratio costs more (29 credits at high) than 16:9, 9:16, 4:3 or 3:4 (20).
Do reference images cost extra?
On OpenAI's API input images are billed as image input tokens ($8.00 per million). On 8frame up to 10 reference images do not change the credit price.
Sources
- OpenAI API pricing (
gpt-image-2image tokens $8.00 input, $30.00 output, $2.00 cached input; text $5.00 input, $1.25 cached; Batch $4.00 / $15.00 / $1.00 and $2.50 / $0.625; "For image generation cost estimates, use the calculator in the image generation guide"): developers.openai.com/api/docs/pricing, checked 2026-10-21 - fal, GPT Image 2 model page (text tokens $5.00 / $1.25 / $10.00, image tokens $8.00 / $2.00 / $30.00 per 1M; per-image table: 1024 x 768 $0.005 / $0.037 / $0.145, 1024 x 1024 $0.006 / $0.053 / $0.211, 2560 x 1440 $0.007 / $0.056 / $0.222, 3840 x 2160 $0.012 / $0.101 / $0.401; "by default we use high"; "Total cost is rounded up to the closest hundredth of a cent"): fal.ai/models/openai/gpt-image-2, checked 2026-10-21
- Replicate, openai/gpt-image-2 (billing tiers in the page's pricing data: low $0.012, medium $0.047, high $0.128, auto $0.128 per output image; quality low / medium / high / auto; bring your own OpenAI API key): replicate.com/openai/gpt-image-2, checked 2026-10-21
- Runway API pricing (
gpt_image_21 / 5 / 20 credits at 1K or 2K and 2 / 11 / 41 at 4K for low / medium / high, "auto" resolution billed at the 4K tier, "Default quality is high"; credits "$0.01 per credit"): docs.dev.runwayml.com/guides/pricing, checked 2026-10-21 - 8frame prices, defaults and limits: the 8frame app, checked 2026-10-21
Pick the quality first and the host second; on GPT Image 2 the first choice moves the price by up to 35 times, the second by up to about four.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.