ChatGPT Image 2 vs 2.5 on Sume: $0.26375 vs $0.065875 at 1024 high
At 1024 square, high quality, ChatGPT Image 2 lists $0.26375 on Sume and Image 2.5 $0.065875, about 4x less. 2.5 adds xhigh, max, 16 references and masks.
On Sume's catalog, ChatGPT Image 2 shows $0.26375 per image and ChatGPT Image 2.5 shows $0.065875, both at 1024 square and high quality, billed. Image 2 costs about four times as much at that setting, so unless an old prompt depends on it, 2.5 is the row to start from.
What the two rows list
| Field | ChatGPT Image 2 | ChatGPT Image 2.5 (Flare, Sunburst) |
|---|---|---|
| API id | openai/gpt-image-2 | openai/gpt-image-2.5, openai/gpt-image-2.5-sunburst |
| Catalog list (1024, high) | $0.211 | $0.0527 |
| Billed (list x 1.25) | $0.26375 | $0.065875 |
| Quality values | low, medium, high | auto, low, medium, high, xhigh, max |
| Reference images | Up to 10 | Up to 16 |
| mask_url and background | Not listed | Listed |
| Aspect ratios | auto, 1:1, 16:9, 9:16, 4:3, 3:4, 5:4, 9:8, 4:5 | The same nine |
What OpenAI says
OpenAI's guide lists the 2.5 variants, Sunburst for editing precision and Flare for fast everyday generation, and says both support six quality levels, while earlier GPT Image models only go up to high (OpenAI guide, read 2026-10-07). That matches Sume's rows: sending xhigh to Image 2 returns a 400.
Why the price gap is not just a discount
The Image 2 figure is a catalog display rate for high quality at 1024. The 2.5 figure comes from fal's token rates ($30 per million output tokens) applied to OpenAI's size and quality calculator. Compare like with like by reading the endpoint's pricing line and your response's usage.cost. At low quality the 2.5 row is far below its high rate: $0.007375 billed at 1024 square, from a $0.0059 list.
Moving a prompt over
- Swap the id to
openai/gpt-image-2.5and keep the same ratio. Both list the same nine values. - Set
qualityyourself. On 2.5, omitted quality ishigh, andautoreserves themaxprice at admission. - Re-run your prompt set. Sume does not claim identical output between the two.
- Use
mask_urlfor single-region fixes, which 2.5 lists and Image 2 does not.
Where Image 2 still fits
Image 2 is still a catalog row. Reasons to keep it are practical: a prompt set that was tuned to it, an approval tied to that look. None of those is a reason to start a new project on it at about four times the price.
If you do keep it, pin the id and put a budget guard around the batch: 1,000 images at 1024 high are $263.75 billed on Image 2 and $65.875 on 2.5. Run a 20-prompt comparison first. At $0.26375 for each image, 20 prompts on Image 2 cost $5.275 against $1.3175 on 2.5, which is cheap compared with the cost of a wrong choice at scale.
Budget check
A 100-image run at 1024 high costs $26.375 on Image 2 and $6.5875 on 2.5, before input tokens on edits. The full request table is on the Image API page.
Sources
Related posts
More in Models
- Do low, medium and high mean the same on every Sume image model?
No. On Sume only ChatGPT Image 2, 2.5 and Ideogram 4.5 accept low, medium and high, and the tiers change price in different ways. Defaults and the tier ladders.
- Does GPT Image 1 still work on October 7, 2026? 16 days left
OpenAI's deprecations page lists the GPT Image 1 shutdown for 2026-10-23, 16 days after 2026-10-07. What still works, what Sume serves and what to move to.
- Gemini Omni video edit on Sume: video_url price by output second
Edit a clip with gemini-omni-flash-1.1 by sending video_url and a prompt. Sume bills $0.125 per second at 720p default, so a 6-second edit is $0.75.
- EmbeddingGemma 2 audio embeddings or a transcript: what to index
EmbeddingGemma 2 embeds audio directly. Sume has no embedding endpoint, but STT gives text and word timings at $0.01 per minute for search.
Written by Sume