OpenAI releases GPT Image 2.5 with two models topping Artificial Analysis leaderboards

OpenAI has released GPT Image 2.5, five months after GPT Image 2, in two variants: Flare and Sunburst. Flare targets everyday text-to-image generation with lower latency than its predecessor. Sunburst targets detailed creative work with tighter edit control, at the cost of longer generation times. Both models support text-to-image and image editing.

On Artificial Analysis's text-to-image and image-editing leaderboards, evaluated at the maximum quality setting, the pair occupy the top two positions. Flare ranks first in text-to-image; Sunburst ranks first in editing.

The largest gap appears in editing. In composition and framing — camera changes, zoom-outs, layout changes that preserve the scene — Sunburst opens the widest margin of any editing action tested. In text and symbol editing — words, numbers, logos, symbols inside an existing image — Sunburst's improvement over GPT Image 2 trails only its composition gain.

Despite the performance gains, estimated output pricing remains unchanged from GPT Image 2 (expensive) for 1024×1024 images at the tested quality settings. Flare, the faster model, generates images in less than half the time of the previous generation, a figure that changes the calculus for latency-sensitive production workloads.

Both models are available now through the OpenAI API. Simultaneously, the company launched ChatGPT Images 2.5 in the standard chat interface, ChatGPT Work, and Codex, giving developers and end users immediate access without waiting for a client-side update.