OpenAI shipped GPT-Image-2.5 on 8 September 2026, rolling out as ChatGPT Images 2.5 across ChatGPT, ChatGPT Work and Codex, plus two new API models called Flare and Sunburst. Generation is up to 50% faster than the previous version and editing holds up better across multiple rounds of changes. The catch: OpenAI priced the API access at exactly double the previous model's rate. If your ad creative, product photography, or content pipeline touches AI image generation, this is worth five minutes today, not a bookmark for later.
What GPT-Image-2.5 Actually Changed on September 8#
OpenAI's own announcement describes two new API models sitting under the GPT-Image-2.5 name. Flare is the default, pitched as producing higher-quality images than GPT-Image-2 at roughly half the latency, aimed at high-volume generation. Sunburst is the slower, tighter-control sibling, built for production creative where you need edits to land exactly where you asked and nowhere else. Both keep earlier edits stable across a long back-and-forth conversation, which older image models were genuinely bad at. OpenAI says more than 3 billion images now get created weekly across ChatGPT Images and the API combined.
At WebEpex we build Meta ad creative and product content for D2C brands and SaaS founders across the GCC, Europe, USA, Canada and India, and this is the exact model swap that decides what our next batch of client product images gets generated on.
Does This Change Anything If You're Not Technical?#
If someone on your team or agency generates product shots, ad variations, or social creative through ChatGPT or an API integration, yes. If you've never touched an AI image tool, this changes nothing about your week.
The people who'll feel it first are the ones running Meta or Google creative through an automated pipeline. n8n workflows, custom scripts, anything wired to the API gets the new pricing whether or not anyone opts in.
The Real Trade-off: Faster, Sharper, Twice the Price#
According to pricing figures published by DataStudios, the new API rates for GPT-Image-2.5 come in exactly double the old GPT-Image-2 rates:
- Image output tokens: $30 per 1M, up from $15
- Image input tokens: $8 per 1M, up from $4
- Text input tokens: $5 per 1M, up from $2.50
That's a real number to plan around if you're generating creative at volume. My read: for a single hero image or a handful of product shots, the price difference is nothing. For an e-commerce catalog running hundreds of SKU images through the API monthly, it's a line item that needs a second look before anyone flips the switch.
The upside isn't just marketing copy. Reference-photo preservation is meaningfully better, and that matters if you're generating variations of an actual product rather than a generic stock-style image. Fewer throwaway generations to get one usable shot can offset a chunk of the price increase. I don't have a clean number for how much it offsets yet, and I'd be guessing if I gave you one.
Who Should Care About This Right Now?#
D2C brands generating product or lifestyle imagery, agencies running Meta ad creative through an automated pipeline, and SaaS teams generating marketing assets in bulk. If that's not you, this is a five-minute read and nothing else.
How We're Already Using It#
At WebEpex we tested Flare against the reference-photo workflow we use for e-commerce catalog work the same day it shipped, including a multi-seller platform build where product photos need to stay visually consistent across hundreds of listings from different sellers.
We swapped our creative-generation step over to Flare by default for that kind of work, and we're keeping Sunburst in reserve for the campaign hero shots that go through a manual review before they touch an ad account.
We didn't touch anything running on GPT-Image-2 mid-flight for existing clients. New generation work starts on 2.5. That's the whole migration, and it took less time to make the call than it took to write this paragraph.
This sits next to two other things worth tracking if AI tooling touches your Meta ad accounts or your content production pipeline: platforms are shipping fast enough right now that a pricing or permissions change can land with a few days' notice, the way WhatsApp's own billing shift did last week.
What We'd Tell a Client This Week#
Don't panic-migrate. If you're generating fewer than a couple hundred images a month, the cost difference is small enough to ignore and the quality jump is worth taking anyway. If you're at real volume, pull last month's image-generation spend, double it, and decide whether Flare's speed and consistency gains are worth that number before your next invoice shows up with the new rate already applied.
If you want a second pair of eyes on what this actually costs for your specific setup, send me what you're running and I'll tell you straight. Takes two minutes, no pitch attached.
