OpenAI Rolls Out Two Image Models: ChatGPT Images 2.5 Splits Into Flare and Sunburst

OpenAI is updating its image generation capabilities with ChatGPT Images 2.5, but the release is not a single model. Instead, the company is splitting the upgrade into two separate systems: Flare, designed for faster generation, and Sunburst, which focuses on more precise edits. The move reflects a growing trend in AI image tools where speed and control are increasingly traded off against each other, and it leaves developers and everyday users wondering which model they will actually encounter in the interface.

According to tests cited by The Decoder, Flare lives up to its name for straightforward prompts, producing images in seconds rather than tens of seconds. Sunburst, on the other hand, shows its strength when users need to modify specific elements of an existing image—changing an object’s color, altering a background, or refining a detail. But the article notes that the assignment of models to users is not transparent, and OpenAI has not clarified the rollout schedule for the ChatGPT interface or the API.

The lack of clarity is not just a minor inconvenience. For developers building on OpenAI’s image API, the difference between Flare and Sunburst can mean reworking prompts, adjusting latency expectations, or even switching models mid-project. A developer who relies on quick iterations for a design tool might prefer Flare, while one building a photo-editing app would need Sunburst’s precision. Without a clear way to select or even identify which model is running, those choices become guesswork.

OpenAI’s approach mirrors a broader industry shift toward specialized models rather than a single do-everything system. Google’s Imagen and Stability AI’s Stable Diffusion have both experimented with separate models for speed and fidelity, and OpenAI’s split suggests the company sees distinct user needs emerging. The trade-off is real: faster generation often means less control over fine details, while precision editing requires more computation time. By offering both under one banner, OpenAI is trying to cover the spectrum without forcing users to learn multiple products.

Early tests hint at who benefits most. Casual users asking for quick illustrations or social media graphics will likely see Flare’s speed as a win, while professional creators—designers, marketers, and artists—will find Sunburst’s editing capabilities more valuable. But the tests also show that results are inconsistent: sometimes the same prompt behaves differently depending on which model is assigned, which could frustrate users who expect predictable outputs.

The bigger question is how OpenAI will communicate this split going forward. If the company keeps the model assignment opaque, users may end up confused about why their results vary. If it offers explicit selection, it could open the door to tiered pricing or usage limits. For now, the release is a reminder that even as AI image quality improves, the user experience depends heavily on how companies package and deploy these models—not just on the underlying technology.

What to watch next: OpenAI has not announced a timeline for broader availability or API access for Flare and Sunburst. Developers and power users should test both models carefully once access expands, and keep an eye on OpenAI’s documentation for any hints about selection logic. Until then, the split remains a work in progress—one that could redefine how people think about image generation speed versus control.