--- title: "GPT Image 2.5: Sunburst vs Flare, the Oct 14 Outage and Credits" description: "GPT Image 2.5 has two models: Sunburst for precise edits, Flare for fast everyday images. A selection table, the Oct 14 incident, and Codex credit costs." url: https://cdnping.org/en/blog/gpt-image-2-5-sunburst-flare language: en published: 2026-10-20 category: "Model watch" tags: ["GPT Image 2.5","OpenAI","image generation","Codex","usage limits"] source: cdnping.org license: Free to quote and summarize with a link to the url above. --- # GPT Image 2.5: Sunburst vs Flare, the Oct 14 Outage and Credits GPT Image 2.5 comes in two models: gpt-image-2.5-sunburst and gpt-image-2.5-flare. According to the OpenAI API changelog, Sunburst suits workflows where editing precision matters most, and Flare suits fast, high-quality everyday generation. On October 14, 2026, the OpenAI status page logged a "limited availability" incident affecting both models. The information here is current as of October 2026; the official pages are authoritative. ## When was GPT Image 2.5 released, and which models are there? Per the [OpenAI API changelog](https://developers.openai.com/api/docs/changelog), both models were released on September 8, 2026. They work with the Image API and the image generation tool in the Responses API, for both generation and editing. The changelog notes: - Sunburst: for workflows where editing precision matters most. - Flare: for fast, high-quality everyday generation. - Both support the new `xhigh` and `max` quality tiers. - Both are billed at GPT Image 2 token rates. The changelog gives no specific numbers, so check the image generation section of the official pricing page. The release is now more than a month old, so two follow-up topics matter more: the October 14 incident and how image generation consumes your usage allowance. ## How should you choose between Sunburst and Flare? OpenAI only gives positioning descriptions, not benchmark numbers. The table below is a selection suggestion based on that positioning, not a measured result. | Scenario | Try first | Reason | | --- | --- | --- | | Local edits on an existing image, high detail requirements | Sunburst | Officially positioned for editing precision | | Batch illustrations, sketches, everyday assets | Flare | Officially positioned for fast everyday generation | | Not sure | Run a few images on each model with the same prompt | Your own asset style is the final judge | | Budget-sensitive | Start with a lower quality tier | `xhigh` and `max` are new tiers; actual cost is on the pricing page | ## What happened on October 14? The [OpenAI status page](https://status.openai.com/history) recorded an incident on October 14 titled "Limited availability on gpt-image-2.5-flare and gpt-image-2.5-sunburst". The page shows 10:58 AM (no time zone is given), the status is resolved, and it states that the affected services have fully recovered. The page lists no specific components and no further detail, so we won't speculate. If your service depends on these models, we suggest: 1. Make the model name a config value instead of hardcoding it. 2. Add a timeout and a bounded number of retries to image requests, and show users a clear error on failure instead of waiting forever. 3. To tell whether the problem is on OpenAI's side or your own network, check the probe results at [/en/status/openai](/en/status/openai) first, then the official status page. For the reasoning, see [Is ChatGPT down or is it my network? A three-step check](/en/blog/is-chatgpt-down-or-my-network) and [How to read the OpenAI and Claude status pages](/en/blog/how-to-read-ai-status-pages). ## How are credits counted for image generation in ChatGPT and Codex? The [ChatGPT pricing documentation](https://learn.chatgpt.com/docs/pricing) says: - Image generation shares the same usage limits as local messages and cloud conversations. - The documentation says image generation consumes included usage at roughly 3–5 times the average rate, and the exact multiple depends on image quality and size. - Once included usage runs out, image generation keeps drawing down credits. - The free plan does not include image generation. - Users with an API key are billed at API prices and do not use up ChatGPT usage limits. For people working in Codex, the implication is simple: a turn with images may exhaust the allowance in your five-hour window faster than a text-only turn of the same length. For how plans and limits are calculated, see [ChatGPT Pro splits into $100/$200/$500 tiers](/en/blog/chatgpt-pro-500-plan-codex-limits). ### Usage self-check list - First confirm whether you are on subscription usage or an API key; they are billed differently. - If you need a lot of images, move generation into a separate script that calls the API, so it doesn't eat the allowance you use for coding. - Use a low quality tier for drafts and raise it for the final version. - Log the model, quality tier and size of every request. Otherwise you won't be able to tell where the money went at month end. ## How can you keep tracking changes like this? For models being added, retired or repriced, use [/en/models](/en/models) to see monitoring records for each vendor, and [/en/changes](/en/changes) to see additions and removals in official changelogs and pricing pages. For a method of comparing prices, see [How to compare LLM API prices](/en/blog/llm-api-pricing-how-to-compare). For tracking changes, see [How to find out right away when an LLM API price changes](/en/blog/track-llm-price-and-model-changes). One more note: we could not verify, in the OpenAI changelog, any entry saying Codex CLI 0.158.0 added transparent-background image support. What the changelog does confirm is that from August 20, 2026, transparent backgrounds for gpt-image-2 are available in preview through the Images API and the Responses API. You need to set `background` to transparent and use PNG or WebP output (JPEG does not support transparency). For the exact behavior inside Codex, rely on the official Codex release notes. ## Sources - [OpenAI API changelog](https://developers.openai.com/api/docs/changelog) - [ChatGPT pricing and usage limits](https://learn.chatgpt.com/docs/pricing) - [OpenAI status page history](https://status.openai.com/history)