OpenAI's GPT Image 2.5 Sunburst trades speed for precision in new image model

OpenAI has released GPT Image 2.5 Sunburst, the precision tier in its new family of image generation and editing models. Unlike its faster siblings, Sunburst targets creative work where granular control over every edit matters more than response time, and it is available through a dedicated Images API.

The model supports a 400-thousand-token context window, enough to stream complex editing instructions, long visual references and extended conversation histories without truncation. Pricing is set at $8 per million tokens for both input and output — roughly 29 shekels per million — placing it at the upper end of commercial image models and reflecting an intent to serve production-grade pipelines rather than casual experimentation.

End-to-end latency measures 21.5 seconds at the median (P50) on the fastest provider, substantially higher than prior-generation models that sometimes dipped below ten seconds. OpenAI reports 100% availability in its internal measurements and 86.26% across external cloud providers, with an automatic failover mechanism that routes to a backup provider when the primary returns an error, provided the request settings allow it.

According to the company, Sunburst is built for production-ready creative and polished product imagery — ad campaigns, commercial catalogs and workflows where every pixel counts. The deliberate slowdown in generation speed is a direct trade-off for tighter editing capability: localized changes, style consistency across a series, and more faithful adherence to complex text instructions.

Sunburst is not a general-purpose model. Teams that need rapid iteration, sketches or high-volume real-time output will find it too slow and too expensive. But for organizations already paying a premium for edit precision, and receiving in return a wide context window, availability with automatic failover and predictable pricing, it is an option worth evaluating in a staging environment before promoting to a live campaign.