GPT Image 2.5 resolution: the size rules and the 4K ceiling
GPT Image 2.5 takes custom sizes: both edges a multiple of 16, longest edge 3840, 655,360 to 8,294,400 pixels. What that means for 1080p, 1440p and 4K.

GPT Image 2.5 accepts custom output sizes inside one box: both edges a multiple of 16, the longer edge at most 3840 pixels, an aspect ratio no wider than 3:1, and a total between 655,360 and 8,294,400 pixels. The top of that box is 3840x2160, which is 4K. There is no fixed list of resolutions to pick from, and no resolution: "4K" tier on these models.
Sume's Image API docs state the same rules for openai/gpt-image-2.5 and openai/gpt-image-2.5-sunburst; OpenAI's image generation guide gives the same limits for the model. Both read 2026-09-29.
What sizes can I actually send?
Set image_size to a named preset, auto, or custom pixels as WIDTHxHEIGHT or { width, height }. Common targets, checked against the rules above:
| Size | Multiples of 16 | Total pixels | Fits the box |
|---|---|---|---|
| 1024x1024 | Yes | 1,048,576 | Yes |
| 1920x1080 | 1080 is not (67.5 x 16) | 2,073,600 | No; use 1920x1088 |
| 2560x1440 | Yes | 3,686,400 | Yes |
| 3840x2160 | 2160 = 135 x 16 | 8,294,400 | Yes, exactly at both ceilings |
| 4096x2304 | Longest edge over 3840 | 9,437,184 | No |
Is 4K reliable?
It is supported, with a caveat from the vendor: OpenAI's guide says resolutions above 2560x1440 are experimental. On Sume, slow configurations such as 4K, high quality and several images per call are the ones most likely to return a 202 job envelope instead of an image, so read the status code, not the body shape.
Can I ask for a ratio instead of pixels?
Yes. aspect_ratio lists auto, 1:1, 16:9, 9:16, 4:3, 3:4, 5:4, 9:8 and 4:5 for these ids. Ratios that need custom pixels are converted to legal ones: 4:5 becomes 1024x1280, because 1080x1350 breaks the multiple-of-16 rule. When you edit a reference image, prefer aspect_ratio: "auto" so the output follows the source; leaving the field out is not the same as sending auto.
What if I need exactly 1920x1080?
These ids cannot produce it directly, since 1080 is not a multiple of 16. Generate at 1920x1088 and trim the extra 8 pixel rows in your own pipeline. Do not send resolution or size: "1080p" for pixel control: Sume's docs say custom pixels belong on image_size, not size.
curl -X POST "https://api.sume.com/v1/images" \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-image-2.5",
"prompt": "wide studio shot of a ceramic mug on a wooden desk",
"image_size": "1920x1088",
"quality": "medium"
}'Sources
Related posts
More in Models
- Hailuo 3.0 release date: MiniMax H3 launched 2026-07-31
Hailuo 3.0 is the name people use for MiniMax H3, which MiniMax released on 2026-07-31. What the vendor pages call it, and the ids that reach it on Sume.
- Imagen 4 API shut down in the Gemini API: what Sume's catalog lists
Google says Imagen is shut down in the Gemini API and points to Nano Banana. Sume's image catalog lists google/imagen-4-fast, imagen-4-ultra and Nano Banana ids
- Kling 2.6 vs 3.0: length, resolution, audio and price
Kling 3.0 makes 3–15 s clips up to 4K with audio and multi-shot; 2.6 makes 5 or 10 s shots at 720p or 1080p. Kling's own specs and prices, dated.
- Kling 3 4K API: Kling's own tier and what Sume lists
Kling 3.0 lists 4K on Kling's own capability map, but Sume's kling-3 lists 720p and 1080p and prices no resolution tier. Sume's 4K route: gemini-omni-flash-1.1.
Written by Sume