What is the cost of image input tokens for the GPT-Image-2.6 API?

Image input tokens cost $8.00 per 1 million tokens, while cached image input tokens cost $2.00 per 1 million tokens.

What are the differences between Sunburst and Flare model variants?

Sunburst handles high-detail image generation and editing with a text-to-image Elo score of 1409. Flare produces faster, lower-latency visual outputs tailored for drafts and product imagery.

What new features are included in GPT-Image-2.6?

New features include animated GIF output from a single prompt, SVG export via the API, markup-guided edits using transparent annotation masks, and adjustable complexity settings.

GPT-Image-2.6 API Pricing and Features Guide

GPT-Image-2.6 API: Pricing & Benchmarks. A macro shot of a dense silicon die reflecting sharp prismatic lens flares and amber light along precision-etched copper interconnects inside a dark server chassis — GPUniq

Data as of November 6, 20266 sources

Drafted with AI tools and edited by a human. The figures and conclusions were checked by a GPUniq editor.

GPT-Image-2.6 API pricing

OpenAI prices the API by token type. Image input tokens cost $8.00 per 1M tokens. Cached image input tokens drop to $2.00 per 1M tokens. Image output tokens cost $30.00 per 1M tokens. Text input tokens are $5.00 per 1M tokens. Cached text input tokens cost $1.25 per 1M tokens.

Model variants

Sunburst handles high-detail image generation and editing. Flare produces faster, lower-latency visual outputs for drafts and product imagery. Both variants are available via the OpenAI Image API for generations and edits endpoints.

Variant Name	Target Workflow
Sunburst	High-detail image generation and editing
Flare	Fast visual generation for drafts and product imagery

Elo scores

Vandata Team benchmarks assign Sunburst an Arena Text-to-Image Elo score of 1409. The Arena Image Edit Elo score for Sunburst reaches 1476.

New features

The model adds animated GIF output from a single prompt and SVG export via the API. Markup-guided edits use transparent annotation masks. Users can also adjust complexity settings to low, medium, or high.

Methodology

Facts are compiled from the vendor's own announcement and documentation, linked below, at the time of writing; where the vendor hasn't published a number it is marked as such. GPUniq prices, when the model is already available here, come from our live catalog.