gpt-image-2-web when the lower USD 0.03 fixed price is the main constraint. Choose gpt-image-2-vip when the application needs the route whose full quality enum has been exercised across the documented workflow.
Decision table
Fixed VIP price versus OpenAI Standard output cost
For a1024x1024 image, OpenAI’s current cost table lists these Standard image-output amounts. OpenAI charges text and source-image input separately, so its column is not a complete request price.
The percentages use
(OpenAI Standard image output - YingTu fixed request) / OpenAI Standard image output, checked September 3, 2026. They do not compare Batch pricing, taxes, promotions, account terms, or total OpenAI input cost. Web uses a different upstream path, so this table does not assign its requests to an OpenAI Standard tier.
When quality changes the decision
VIP quality values control the output image budget. Uselow for draft-oriented work, medium for a general default, high when the output needs the largest supported budget, or auto to let the route select. Invalid values fail before generation.
Quality is not a latency guarantee. A higher tier may consume a larger image token budget without completing in a predictable time order.
Web forwards quality, but the selected upstream determines its effect. Do not use an HTTP 200 alone as evidence that two Web quality values produce distinct budgets or visual outcomes.
What remains the same
Both models share Bearer authentication, synchronous Generate and Edit paths, the unified/v1/images/tasks endpoint, 1–16 asynchronous reference images, task states, the three documented dimensions, quality forwarding, and one returned image. Switching normally requires changing only model.
Web parameters
Copy the lower fixed-price request and supported quality values.
VIP parameters
Select a quality and size safely.