Is ChatGPT Images 2.5 free?

Yes. OpenAI rolled Images 2.5 out to all ChatGPT tiers on 8 September 2026, including free accounts, as well as ChatGPT Work and Codex. Free accounts still generate fewer images per day than paid ones, so the model is free to use but the number of images you get is not unlimited.

How do I use Sketch in ChatGPT?

Type @Sketch in a chat to open a drawing canvas inside ChatGPT. Draw a rough layout, then describe what you want, and the model uses your drawing as a reference for the final image. If typing @Sketch does nothing, Images 2.5 has not reached your account yet.

What is the difference between GPT-Image-2.5 Flare and Sunburst?

Both are API models released alongside Images 2.5. Flare is the general-purpose option and carries the full speed and quality improvements. Sunburst takes longer to generate but offers more precision for detailed creative work. They are priced identically.

Why am I hitting an image limit on ChatGPT Plus?

OpenAI applies daily image generation caps by tier and did not raise or publish new numbers with the Images 2.5 launch. Users on paid tiers reported hitting caps on launch day, most likely because of the volume of people testing the new model. The cap resets on a rolling daily basis.

Do I need to update the ChatGPT app to get Images 2.5?

No. This is a server-side rollout, so the new model appears on its own with no app update required on iPhone, Android, desktop or the web. If you do not have it yet, reinstalling or force-closing the app will not make it arrive sooner.

ChatGPT Images 2.5 Rolls Out With Sketch, 50% Faster

What is actually new

The headline change is speed. OpenAI puts the latency cut at up to half compared with the previous generation, which matters most when you are iterating on the same picture ten times rather than generating one and walking away.

The quality claims are more specific than usual. The model is meant to render more natural lighting and richer surface texture, and to hold onto the subject of a reference photo instead of quietly redrawing someone's face between edits. It is also supposed to follow instructions consistently across several turns of the same conversation, which is where earlier image models tended to drift.

Sketch and Templates

Two interface additions ship alongside the model:

Sketch lets you draw inside ChatGPT and hand that drawing to the model as a layout reference. Type @Sketch in a chat to open it.
Templates are preset starting points for common formats — posters, merch designs, flyers, product shots — so you are not writing a prompt from a blank box.

There is also a comment-based editing flow: you annotate a spot on a generated image and the model changes only that region, rather than regenerating the whole frame and hoping the rest survives.

Who gets it, and where

This is a worldwide rollout with no market restriction attached. It covers ChatGPT, ChatGPT Work and Codex accounts across desktop, mobile and the web, on all subscription tiers including free. That is a change of pattern for OpenAI, which has spent the past year gating new models behind Plus and Pro first — GPT-6 Astra landed for paid tiers only before free users saw it.

What the announcement does not do is publish per-tier image quotas. Free accounts have always generated fewer images per day than Plus or Pro, and that has not visibly changed here; discussion threads on launch day already included people on Plus reporting they had hit a daily cap. If you are on the free tier and images stop generating, a quota is the likely explanation rather than a fault.

Two new models for developers

The API side gets two variants rather than one. GPT-Image-2.5 Flare is the everyday option, carrying the same quality and speed gains as the ChatGPT-facing model. GPT-Image-2.5 Sunburst trades generation time for precision, aimed at detailed creative work where an extra few seconds is worth it.

According to published pricing for the pair, both are billed identically: $8.00 per million image input tokens, dropping to $2.00 for cached input, $30.00 per million image output tokens, and $5.00 per million text input tokens, or $1.25 cached.

How to check which model you have

There is no app update to install — this is a server-side change, so it appears on its own. To confirm you are on the new version, open a chat and type @Sketch. If the drawing canvas opens, your account has Images 2.5; if nothing happens, the rollout has not reached you yet and force-closing the app will not speed it up.

Sketch and Templates are additions to the existing image flow, not replacements, so prompting an image the way you always have still works. For anyone comparing image tools, Google's rival stack moved recently too, with Gemini Omni 1.1 Flash extending AI video to 40 seconds.