gpt-image-2 or gpt-image-2.5 to replace gpt-image-1 on Sume?

Pin openai/gpt-image-2.5-sunburst or openai/gpt-image-2.5 for a gpt-image-1 replacement, not openai/gpt-image-2. That is what OpenAI's deprecations page names for gpt-image-1, gpt-image-1-mini and gpt-image-1.5 as of 2026-10-06. gpt-image-2 appears on the same page only as a replacement for the retired DALL-E ids.

On Sume the practical difference is in what the request may carry. Only the 2.5 pair takes mask_url, background, up to 16 input references, and the xhigh and max quality tiers.

Request differences on Sume

What happens if you send an unsupported field

Sume does not drop a field the model does not list. It returns 400 unsupported_parameter. If you call gpt-image-2 with mask_url, you get the 400 and the edit never runs. The sample picks the model from the fields the request needs:

def pick_model(body):
    needs_25 = (
        "mask_url" in body
        or "background" in body
        or len(body.get("input_references", [])) > 10
        or body.get("quality") in ("xhigh", "max")
    )
    body["model"] = ("openai/gpt-image-2.5-sunburst" if needs_25
                     else "openai/gpt-image-2")
    return body

if __name__ == "__main__":
    print(pick_model({"prompt": "a mug", "mask_url": "https://example.com/m.png"}))
    print(pick_model({"prompt": "a mug"}))

Which should be your default?

Make 2.5 the default if you want one id to carry every old gpt-image-1 call, including edits. Sunburst and Flare share price and limits, as covered in the Flare versus Sunburst post. Run your own prompts on both before you commit; Sume's catalog can tell you what each accepts, but not which looks better for your product. See the 2.5 API guide.