GPT Image 2.6 is billed by tokens, not per image, so the price of one picture depends on three settings: quality, size and shape. OpenAI charges $30 per 1M image output tokens for both variants, Sunburst and Flare. We ran 51 paid requests on November 10, 2026 and recorded the token usage returned for each. A 1:1 image at 1K used 824 output tokens at low quality and 7,024 at max, which is about $0.025 to $0.21 per image at OpenAI's rate. Size and shape move the number as much as quality does: 4K costs 3.4 times 1K, and a 3:1 panorama costs a third of a square.
Last verified: 2026-11-10.
Quick Answer
| 1:1 image at 1K | Output tokens | At $30 per 1M tokens | Credits on SeedVideo AI |
|---|---|---|---|
| low | 824 | $0.025 | 2 |
| medium | 1,449 | $0.043 | 3 |
| high | 1,756 | $0.053 | 3 |
| xhigh | 3,122 | $0.094 | 5 |
| max | 7,024 | $0.211 | 10 |
Low and medium at 1:1 are our own measurements. For high, xhigh and max we measured a 3:4 image (table below) and the 1:1 values are the ones the provider documents, which our 3:4 numbers line up with.
TL;DR
- Sunburst and Flare cost the same. They returned identical token counts at every setting we tried.
- Medium and high are close (1,449 and 1,756 tokens). Max is 4.8 times medium.
- 2K is about 2 times the tokens of 1K, and 4K about 3.4 times.
- Wider or taller shapes are cheaper: tokens scale with the short side divided by the long side.
- Compared with GPT Image 2.5, low and medium cost 3 to 4 times more tokens. High, xhigh and max are unchanged.
- The complexity setting and a transparent background did not change the token count.
The official rates
The OpenAI model pages for GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare list the same prices:
| Token type | Per 1M tokens | Cached input |
|---|---|---|
| Text input | $5.00 | $1.25 |
| Image input | $8.00 | $2.00 |
| Image output | $30.00 | — |
Output tokens are nearly the whole bill. Our prompts were 53 to 128 text tokens, which is less than a tenth of a cent. Resellers and API gateways may charge a different rate per token, but the token counts below are what the model itself reports.
Quality: five tiers on the same image
We ran one chalkboard-menu prompt at 3:4 and 1K (880 × 1184) on every tier, on both variants.
GPT Image 2.6 Sunburst, left to right: low, medium, high, xhigh, max.
GPT Image 2.6 Flare, left to right: low, medium, high, xhigh, max.
| Quality (3:4, 1K) | Output tokens, Sunburst | Output tokens, Flare | At $30 per 1M | Time, Sunburst | Time, Flare |
|---|---|---|---|---|---|
| low | 618 | 618 | $0.019 | 23 s | 108 s |
| medium | 1,086 | 1,086 | $0.033 | 346 s | 26 s |
| high | 1,315 | 1,317 | $0.039 | 58 s | 277 s |
| xhigh | 2,337 | 2,337 | $0.070 | 121 s | 175 s |
| max | 5,184 | 5,184 | $0.156 | 128 s | 83 s |
All ten images spelled every word and price correctly. On this prompt, paying for a higher tier bought finer chalk texture and lighting, not accuracy. Max cost eight times as much as low for a result most people would struggle to tell apart at this size.
The times are single runs and include queue time. They do not follow the tiers: Sunburst's slowest was medium and Flare's was high. In launch week, waiting was driven more by load than by the quality setting. Our repeated timing runs are in the GPT Image 2.6 vs 2.5 test.
Size: 1K, 2K and 4K
GPT Image 2.6 Sunburst at medium quality: 1K, 2K and 4K. Each is a separate generation of the same prompt.
| 1:1 at medium quality | Pixels | Output tokens | At $30 per 1M | Compared with 1K |
|---|---|---|---|---|
| 1K | 1024 × 1024 | 1,449 | $0.043 | 1× |
| 2K | 2048 × 2048 | 2,944 | $0.088 | 2.03× |
| 4K | 2880 × 2880 | 4,894 | $0.147 | 3.38× |
A 2K image has four times the pixels of a 1K one but only twice the tokens, so higher resolution is cheaper per pixel. Note that 4K at 1:1 is 2880 × 2880, not 4096: the model caps an image at 8.29 megapixels and 3,840 pixels per edge.
Shape: wide and tall images cost less
| Medium quality at 1K | Pixels | Output tokens | Share of the 1:1 figure |
|---|---|---|---|
| 1:1 | 1024 × 1024 | 1,449 | 100% |
| 4:5 | 912 × 1152 | 1,149 | 79% |
| 3:4 | 880 × 1184 | 1,086 | 75% |
| 3:1 | 1776 × 592 | 486 | 34% |
All four images have about the same number of pixels, yet the panorama used a third of the tokens. In every case the count is close to the square figure multiplied by the short side divided by the long side (4 ÷ 5, 3 ÷ 4, 1 ÷ 3). The same held at 2K: a 16:9 image at high quality used 2,010 tokens, against roughly 3,560 for a square one.
A practical rule for estimating any image: take the 1:1 tokens for the quality tier, multiply by 1, 2.03 or 3.38 for 1K, 2K or 4K, then multiply by short side ÷ long side.
Edits, GIFs and transparency
Edits. Each reference image is billed as image input. Our 1024 × 1024 reference added 1,024 input tokens, about $0.008, on top of the normal output tokens. The output cost did not change.
GIFs. A GIF is billed on the sheet of frames the model draws, not per frame. Our 9-frame, 512 × 512 loop used 1,172 output tokens on Flare at low quality ($0.035) and 2,073 on Sunburst at medium ($0.062). That is more than a single 1K image at the same tier, but far less than nine images.
GPT Image 2.6 Sunburst, medium quality, 9 frames at 120 ms: 2,073 output tokens, 28 seconds.
GPT Image 2.6 Flare, low quality, same prompt: 1,172 output tokens, 14 seconds. Flare drew motion lines we did not ask for.
Transparent background. Our sticker test returned a PNG with a real alpha channel and used 1,449 tokens, the same as an opaque 1:1 medium image.
Complexity. We ran the same infographic at low, medium and high complexity. Each used 1,149 tokens. The setting changes how long the model plans, not what you pay.
GPT Image 2.6 vs 2.5: what changed in price
| Quality (1:1 at 1K) | GPT Image 2.5 output tokens | GPT Image 2.6 output tokens | Change |
|---|---|---|---|
| low | 196 | 824 | 4.2× |
| medium | 439 | 1,449 | 3.3× |
| high | 1,756 | 1,756 | Same |
| xhigh | 3,122 | 3,122 | Same |
| max | 7,024 | 7,024 | Same |
We measured the medium row on both models with the same prompt (439 and 1,449 tokens); the other 2.5 values are from the provider's documentation. The rate per token did not change between generations, so the increase is entirely in how many tokens the two lowest tiers use. If you generate drafts in volume, GPT Image 2.5 is still the cheaper model.
What it costs on SeedVideo AI
SeedVideo AI charges credits per image. The cost depends on quality and size, not on aspect ratio, and it is the same for Flare and Sunburst.
| Quality | 1K | 2K | 4K |
|---|---|---|---|
| low | 2 credits | 3 credits | 4 credits |
| medium | 3 credits | 5 credits | 7 credits |
| high | 3 credits | 5 credits | 9 credits |
| xhigh | 5 credits | 9 credits | 15 credits |
| max | 10 credits | 20 credits | 34 credits |
High costs the same as medium at 1K and 2K, so there is no reason to stop at medium there. Reference images add to the cost: a medium 1K image stays at 3 credits with one or two references and reaches 10 credits with 16. The generator shows the exact number before you run a prompt. GIF output, markup edits and the complexity setting are in the generator too: a GIF starts at 1 credit at low quality (2 credits for nine frames), and markup and complexity add nothing to the cost. The GPT Image 2.6 page shows examples of each.
Plans and credit packs are on the SeedVideo AI pricing page, and you can run the model on the GPT Image 2.6 page.
How to spend less
- Draft at low or medium, 1K. Every tier spelled our text correctly, so start cheap and move up only for texture.
- Use the shape you need. A 16:9 or 3:1 image costs far fewer tokens than a square of the same pixel count.
- Go to 2K or 4K only for the approved prompt. Rerun the final wording once at the delivery size.
- Keep GPT Image 2.5 for volume. Below high quality it uses a quarter to a third of the tokens.
- Skip max unless you can see the difference. It is 4.8 times the tokens of medium.
FAQ
How much does one GPT Image 2.6 image cost?
At OpenAI's $30 per 1M output tokens, a 1:1 image at 1K costs about $0.025 at low, $0.043 at medium, $0.053 at high, $0.094 at xhigh and $0.21 at max. Larger sizes cost more and elongated shapes cost less.
Is Flare cheaper than Sunburst?
No. Both have the same token rates and used the same number of output tokens at every setting in our test. Flare is faster, not cheaper.
Why is GPT Image 2.6 more expensive than 2.5?
The per-token rate is the same, but 2.6 uses more output tokens at low (824 against 196) and medium (1,449 against 439). From high quality upward the two are identical.
How much does a 4K image cost?
A 1:1 medium image at 4K (2880 × 2880) used 4,894 output tokens, about $0.147 at OpenAI's rate, or 7 credits on SeedVideo AI.
Do GIFs cost more than still images?
Somewhat. Our 9-frame, 512 × 512 GIF used 1,172 output tokens at low quality, compared with 824 for a single 1K image at the same tier.
Does the complexity setting cost extra?
No. Low, medium and high complexity returned the same token count for the same prompt.
