OpenAI names Flare or Sunburst to replace gpt-image-1: which to pin?

OpenAI's deprecations page does not pick one for you: the replacement column for gpt-image-1 reads gpt-image-2.5-sunburst or gpt-image-2.5-flare. On Sume both are in the Image API catalog, openai/gpt-image-2.5 for Flare and openai/gpt-image-2.5-sunburst for Sunburst, with the same price, the same quality tiers and the same limits. Pin Flare if you want OpenAI's "fast, high-quality everyday image generation" for volume, and Sunburst where you want what OpenAI calls its "most capable model for image generation and editing". Because cost does not separate them, run a 20-prompt A/B on your own prompts and let the images decide.

The shutdown date to plan around is 2026-10-23 for gpt-image-1. That is 16 days after this post's read date.

What OpenAI's page says, dated

All three retiring OpenAI image ids point at the same two replacements. The rows below are from the deprecations page; the model pages for Flare and Sunburst list identical token rates, so OpenAI is not pricing one above the other either.

What is identical on Sume

Both Sume ids accept text-to-image, up to 16 reference images through input_references, an optional mask_url, and background: auto|transparent|opaque. The quality field takes auto, low, medium, high, xhigh or max, and Sume's default if you omit it is high. image_size takes named presets, auto, or custom pixels where both edges are multiples of 16, the longest edge is at most 3840, the ratio is at most 3:1 and the area is between 655,360 and 8,294,400 pixels.

Pricing is the same token math for both: $30 per million output image tokens, $8 per million input image tokens and $5 per million input text tokens, rounded up to $0.0001, then multiplied by 1.25 for the billed amount. At 1024x1024 that is $0.0132 list and $0.0165 billed at medium, $0.0527 list and $0.065875 billed at high.

A script that settles it

Write one prompt set that looks like your real workload, include at least two with text inside the image, and send each prompt to both ids at the quality you intend to ship. The loop below prints the billed cost and the hosted URL for each. POST /v1/images waits up to 30 seconds and returns 200; a slower job comes back as 202 with a job envelope, which the script prints so you can poll it through the job result endpoint.

import os
import requests

URL = "https://api.sume.com/v1/images"
HEADERS = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
PROMPT = "A matte black water bottle on a stone ledge, morning light"

for model in ("openai/gpt-image-2.5", "openai/gpt-image-2.5-sunburst"):
    body = {"model": model, "prompt": PROMPT, "quality": "medium"}
    r = requests.post(URL, headers=HEADERS, json=body, timeout=60)
    if r.status_code == 200:
        data = r.json()
        print(model, data["usage"]["cost"], data["data"][0]["url"])
    else:
        print(model, r.status_code, "poll the job envelope:", r.text[:200])

How to decide and what to put in config

Put the model id in one config value so the choice is a one-line change, and keep the quality in a second value. A sensible default is Flare at medium for drafts and bulk work, with Sunburst at high for finals that carry dense text or product detail. If your A/B shows no visible difference on your prompts, stay on one id and stop paying attention to the other, since the invoice will not move.

Budget the switch as well as the model. Sixty 1024x1024 images at medium are 60 times $0.0165, or $0.99 billed; at high they are 60 times $0.065875, or $3.95. Those are small numbers, so spend them on a real A/B before you commit your whole library to one id rather than guessing from a spec sheet.

Do not rely on a name appearing in response bodies to tell the two apart. The model field echoes the id you requested, so log the id you sent next to each result yourself.

What not to assume

This post claims only that OpenAI lists both ids as replacements and that Sume serves both at equal price. It does not claim one beats the other, and Sume publishes no benchmark that ranks them. The difference between Flare and Sunburst, as far as OpenAI's pages describe it, is positioning, not price.

Also check smaller ids you may still call. gpt-image-1-mini and chatgpt-image-latest keep working until 2026-12-01 on OpenAI's side, but a migration done once for all three is cheaper than three migrations.