OpenAI · Image · Updated September 14, 2026New: released 8 Sep 2026

GPT Image 2.5: the reference-faithful image model, in two speeds

OpenAI's GPT Image 2.5 shipped on 8 September 2026 as ChatGPT Images 2.5 in the app and as two API models: Flare, the fast default, and Sunburst, the premium model for tighter control across edits. It keeps GPT Image 2's token prices, cuts latency by up to 50%, and holds your product and people from reference photos far better. It makes stills, not video, and here is what that means for a short-form calendar.

What GPT Image 2.5 is

GPT Image 2.5 is OpenAI's image generation and editing model, released on 8 September 2026 in ChatGPT for every tier, in Codex, and in the API as gpt-image-2.5-flare (the default, higher quality than GPT Image 2 at 50% lower latency) and gpt-image-2.5-sunburst (the premium model built for tighter control across successive edits). It generates and edits still images from text and image inputs with inpainting, six quality tiers from low to max, and 2K output, at unchanged token rates of $5 per million text input, $8 per million image input and $30 per million image output tokens. It does not generate video.

For marketers the point is reference fidelity at speed: one product photo becomes a batch of consistent slideshow frames, thumbnails and meme templates without a retouch pass.

GPT Image 2.5 specs

Maker	OpenAI
Released	8 September 2026. Default API snapshots gpt-image-2.5-flare-2026-09-08 and gpt-image-2.5-sunburst-2026-09-08. Supersedes ChatGPT Images 2.0 / GPT Image 2
Type	Image only; no video generation
Models	Flare (fast default for creator content, social media, product experiences and high-volume generation) and Sunburst (premium, slower, built for tighter control across edits)
Modes	Text-to-image, image editing with reference images, inpainting. Endpoints: images/generations and images/edits
Quality tiers	low, medium, high, xhigh, max, auto
Resolution	2K output retained from Images 2.0 with additional supported sizes; the exact size list is not published on the model cards. Early testing measured roughly 1672 x 941 in ChatGPT
Consistency	Improved preservation of people and products from reference photos; one early review found stronger multi-angle character consistency than Nano Banana 2
Speed	Up to 50% lower latency than GPT Image 2 (Flare)
Pricing (API, both models)	$5 per 1M text input tokens ($1.25 cached), $8 per 1M image input tokens ($2 cached), $30 per 1M image output tokens. No flat per-image price; cost depends on quality and size
Rate limits	Tier 1: 100K TPM, 5 images per minute. Tier 5: 8M TPM, 250 images per minute
Watermark	Provenance metadata on 2.5 output not verified on the model cards

What changed in GPT Image 2.5

Versus GPT Image 2 (ChatGPT Images 2.0): sharper detail with more natural lighting and texture, better preservation of people and products supplied as references, more reliable instruction following across multi-turn edits, generation up to 50% faster, and more supported output sizes. The API split into two named models, Flare and Sunburst, is new; per-token pricing did not change. The ChatGPT app also gained a Sketch tool, format templates, comment-based editing and prompt sharing.

How to get GPT Image 2.5

ChatGPT, ChatGPT Work and Codex across all tiers on desktop, mobile and web, with the usual per-plan generation limits; developers use gpt-image-2.5-flare or gpt-image-2.5-sunburst on the OpenAI Images API.

What GPT Image 2.5 means for short-form marketers

The change that matters for a brand account is reference fidelity. Short-form product content only works if the product looks like the product across dozens of frames, and GPT Image 2 drifted enough on references that anything with a logo or a label needed a manual check. GPT Image 2.5 holds the supplied photo far better and keeps holding it through the third and fourth edit, which turns one product shot into a week of consistent slideshow frames.

The Flare and Sunburst split maps neatly onto how a short-form team actually works: Flare for the volume (slide frames, meme templates, thumbnail variants), Sunburst for the one hero frame that becomes a cover or a paid creative. Same endpoints, same pricing per token, one parameter to switch.

It is image-only, and OpenAI is shutting down its Sora 2 video API on 24 September 2026 with no announced replacement, so this is not a video strategy. Pair it with a video model for clips and use it for the image-native formats that TikTok and Instagram still reward.

Where GPT Image 2.5 still needs care

Still images only; no video mode, and OpenAI has no announced successor to the Sora 2 API
Output in ChatGPT measured well under 2K in early testing, so print or large-format use needs an upscale
Complex multi-element prompts still compound errors: stray objects and stretched anatomy appear as scenes get busier
A slightly composited look on foreground subjects is reduced but not gone
No published per-image price; cost is only known from token usage after generation

Using GPT Image 2.5 with Fastlane

Generate anywhere, publish natively to TikTok, Reels, Shorts and LinkedIn, attribute signups and sales to every post.

Generate product frames, meme templates and cover images with GPT Image 2.5 Flare, then build them into Fastlane slideshows and memes, schedule weeks ahead and publish natively to TikTok, Instagram Reels and YouTube Shorts with signups and sales attributed to each post.

Fastlane covers the video half from your website URL: AI UGC videos with 1,000+ hyper-realistic characters, hook + demo videos and trend remixes. Blitz mode lets you swipe through the generated variations. Free plan, no credit card. Read our full GPT Image 2.5 guide for the workflow in detail.

Try Fastlane free See the AI TikTok video generator

GPT Image 2.5: frequently asked questions

What is GPT Image 2.5?

OpenAI's image generation and editing model, released 8 September 2026 as ChatGPT Images 2.5 in the app and as two API models, Flare and Sunburst.

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

Flare is the fast default with higher quality than GPT Image 2 at 50% lower latency, aimed at creator and social content at volume. Sunburst is the premium model built for tighter control across successive edits, aimed at campaign creative and polished product imagery.

How much does GPT Image 2.5 cost?

Token rates are unchanged from GPT Image 2: $5 per million text input, $8 per million image input and $30 per million image output tokens on both models. There is no flat per-image price.

Does GPT Image 2.5 make video?

No. It is image-only. OpenAI is shutting down the Sora 2 video API on 24 September 2026 and has named no replacement.

Is GPT Image 2.5 free?

ChatGPT Images 2.5 is rolling out to ChatGPT, ChatGPT Work and Codex users across all tiers, including free, with per-plan generation limits. API use is billed per token.

Sources

Specs verified against the sources above on September 14, 2026. Model access and pricing change often; check the maker's page before you build on any number here.

Ship 30 days of viral content in 30 seconds

Join 50,000+ founders and marketers publishing the same video to TikTok, Instagram, YouTube and LinkedIn. Free to start.

Try Fastlane free See pricing