OpenAI's GPT Image 2.5 Flare and Sunburst land on Vercel's AI Gateway

What happened

Vercel has added two new OpenAI image models, GPT Image 2.5 Flare and GPT Image 2.5 Sunburst, to its AI Gateway, according to a post on the Vercel blog dated September 9, 2026. Developers can start calling them through the Gateway's unified API under the identifiers openai/gpt-image-2.5-flare and openai/gpt-image-2.5-sunburst, or experiment with them in Vercel's model playground.

Speed or precision

According to Vercel, both models generate and edit images with more natural-looking lighting and textures, can follow involved visual instructions, and cope with detailed layouts and transparent backgrounds. They also accept reference images for targeted edits, changing only what the instruction specifies while leaving the rest of the picture untouched.

The split between the two is a trade-off. Flare is tuned for quick generation and rapid iteration, which makes it the practical choice when a team needs to cycle through drafts or produce variations at volume. Sunburst takes longer to return a result but gives finer control over composition and detail, which Vercel frames as the better fit for careful creative work.

In practice, switching between the two is just a change of model string. Vercel's examples use the generateImage function from its AI SDK: a plain text prompt for generation, and for editing, a prompt object that pairs the instruction with one or more input images read from disk — for instance, telling the model to swap a bottle cap for a gold one while leaving everything else unchanged.

What the Gateway layer adds

The AI Gateway itself is a big part of the story. Vercel describes it as a single API for calling models, with built-in usage and cost tracking, retries, failover, and what the company says are performance optimizations that deliver higher availability than the underlying providers offer on their own. It also ships with custom reporting, budgets scoped to API keys, and routing rules.

Pricing may be the more consequential detail for cost-sensitive teams: Vercel says the Gateway mirrors provider pricing with no markup and charges no platform fee on inference — including on requests made under the bring-your-own-key model, where customers supply their own provider credentials.

Why it matters

Image generation is shifting from a novelty feature to a routine building block in web applications, and access patterns matter as much as raw model quality. By putting OpenAI's newest image models behind one gateway alongside the other models it carries, Vercel lowers the integration cost for developers already building on its platform: a single API, plus shared tooling for spend limits, observability and failover.

The two-model split also reflects a maturing pattern in generative image tooling. Rather than one model trying to serve every use case, vendors are shipping a fast lane for iteration and a slow lane for finish work, and letting developers route per request. That granularity is only useful if switching is cheap — and here it amounts to changing one string. Combined with no-markup pricing and BYOK support, the launch makes it notably frictionless to wire OpenAI's latest image models into production through Vercel.