GPT Image 2.5 API Pricing: Tokens, Quality and Outputs
GPT Image 2.5 is not a universal fixed-price image endpoint on Siray. The current model-detail Pricing sections explicitly calculate actual charges from token usage. A displayed per-image estimate is therefore a starting reference, not a promise that every quality setting, resolution or edit costs the same amount. Budget from the billing rules before multiplying a headline rate by your planned image count.
Key TakeawaysOn October 15, 2026, the four GPT Image 2.5 model pages showed matching token rates, with a 15% discount badge.Matching rates do not guarantee matching bills: input and output token usage still matter.Choose the endpoint, quality and output size before comparing estimates. Conversion examples are not invoices or proof of a permanent price.
What Are the Current Token Rates?
On October 15, 2026, Siray's Pricing blocks for Flare t2i, Flare edit, Sunburst t2i and Sunburst edit each displayed these fees:
| Usage category | USD per 1,000 tokens |
|---|---|
| Image input | $0.0068 |
| Other input | $0.00425 |
| Output | $0.0255 |
The pages state that generated images are approximately $0.003 each, followed by the token rates and an explicit usage-based billing statement. Read those statements together. The approximation is not a flat charge covering every possible request.
At those displayed rates, the usage-based calculation is:
cost = image_input_tokens / 1000 × 0.0068 + other_input_tokens / 1000 × 0.00425 + output_tokens / 1000 × 0.0255
This formula uses the three published categories. It does not estimate how many tokens your particular request will consume.
Why Does the Per-Image Estimate Change?
Quality and resolution appear explicitly in the console's conversion table. On the same check date, Flare t2i displayed the following discounted examples. Keep each value attached to its settings instead of extracting the cheapest number and applying it to every image.
| Quality | Resolution | Displayed cost per run |
|---|---|---|
| low | 1024x768 | $0.003 |
| high | 1024x1024 | $0.045 |
| max | 3840x2160 | $0.34 |
These are console conversion examples, not charges observed from executed requests. No paid generation was run for this refresh. Treat the table as a planning aid, then reconcile actual usage and charges when testing your workload.
Do Editing and Generation Cost the Same?
The four checked pages publish the same token rates, but that does not establish identical total costs. An edit introduces reference images; the resulting input and output usage determines the bill. “Same rate schedule” is narrower and more accurate than “every edit costs the same as every generated image.”
Likewise, matching Flare and Sunburst request fields do not prove identical speed or visual quality. Compare actual outputs for your task before making that choice. The earlier flat $0.004-per-image shortcut and fixed cross-generation savings percentages should not be used for current budgeting.
Which Output Choices Should You Record?
The current Flare t2i specification exposes quality values low, medium, high, xhigh and max. Output formats are jpg, png and webp; background choices are auto, opaque and transparent. Their presence is not a promise that every combination is appropriate for every task.
The Flare edit specification additionally requires an images array, accepting one to sixteen image strings, and offers an optional mask. Both modes require size; n accepts one to ten outputs, defaulting to one. Neither schema exposes seed.
A resolution appearing in a billing table is not automatically a named size preset. Check the schema before constructing a request, including the conditions for custom dimensions. Do not copy a conversion-table label into the API and assume it is accepted.
Can a discount be treated as permanent? No. Recheck the model's Pricing block before quoting or budgeting a new batch.
What should a budget record contain? Endpoint, quality, size, requested output count, reference inputs, applicable rates and observed usage. A model name and image count alone are insufficient.
Review the current GPT Image 2.5 billing details on Siray before committing a production workload.
Sources: Siray model-detail Pricing blocks and OpenAPI specifications linked above, retrieved October 15, 2026. This is an offline refresh draft; no live article or billing account was changed.