ChatGPT Images 2.5 Launches: 50% Faster Image Generation, and Editing Finally Lets You Point and Fix

动察 Beating AI Flash: OpenAI has released ChatGPT Images 2.5. Compared to the previous generation, the new version cuts image generation latency by up to 50%, and when iterating on existing photos, it's now easier to preserve the likeness and details of people and pets.

Editing has also improved this time. Users can directly circle a specific spot in an image and tell ChatGPT what to change—for example, swapping out just one petal or altering the color of a single object. Across multiple rounds of edits, it's also less likely to unintentionally modify other areas.

ChatGPT now also supports direct sketch-based control over composition, and has added common image templates for posters, merchandise, and more.

Images 2.5 is now rolling out to ChatGPT, ChatGPT Work, and Codex users.

The API simultaneously opens up two versions: Flare and Sunburst. Flare is faster, suited for high-volume image generation; Sunburst is more refined, ideal for scenarios demanding higher visual quality.

In the latest Arena evaluations, Sunburst ranks first and Flare second, with both models sweeping the top two spots across text-to-image, single-image editing, and multi-image editing leaderboards.