ChatGPT Images 2.5: Faster, more precise, but not the same for everyone
OpenAI has announced ChatGPT Images 2.5, an update to its image generation capabilities inside ChatGPT that splits functionality across two separate models. Rather than a single unified upgrade, the release introduces Flare, aimed at speed, and Sunburst, aimed at editing precision. The distinction reflects a deliberate architectural choice - optimizing for two different use cases that previously had to be balanced within a single model.
Sunburst appears to be the more technically significant of the two, targeting the kind of fine-grained image editing that has historically been a weak point for generative image tools. Precise edits - changing a specific element in an image without disturbing surrounding details - require a different set of trade-offs than fast generation from a prompt, which may explain why OpenAI chose to separate the two workloads into dedicated models.
What remains less clear is how ChatGPT users will actually encounter these models. OpenAI has not fully specified which model a given user receives, under what subscription tier, or on what timeline. This kind of staged or segmented rollout is not unusual for OpenAI, but it does mean the practical experience of Images 2.5 will vary considerably depending on who is using it and when. Early hands-on testing from The Decoder offers some initial signal on where the improvements are most noticeable, though a complete picture will take time to emerge as access broadens.
The update fits into a broader pattern of OpenAI iterating on its image capabilities since the widely discussed launch of its GPT-4o-based image generation earlier in 2025. That release drew significant attention for its ability to follow complex instructions and render text accurately. Images 2.5 appears to be a refinement pass - improving throughput and edit quality rather than introducing a fundamentally different approach. For users who rely on ChatGPT for image work, the key question will be whether the model they actually receive reflects the improvements being described.
