Image API and Responses image-generation tool behavior can differ. Record quality setting, output size, edits and total billing; a text-model token estimate is insufficient.
The provider positioning is a hypothesis to test. Compare preservation of unchanged image regions and instruction compliance at the same output size and quality setting.
Both GPT Image 2.5 models launched September 8, 2026. OpenAI positions Sunburst for editing precision and Flare for everyday generation. Use separate generation and edit suites to test that positioning.
Use two suites with different success criteria
Generation cases should score composition, prompt adherence and rendered text. Editing cases should also score preservation of the original image. Combining them into one average can hide a model's important failure mode.
Record API and quality settings
Both models are announced through the Image API and Responses image-generation tool, with xhigh and max quality settings. Use the same endpoint and supported output settings, then compare accepted images and actual billed usage.
Three criteria for this comparison
Generation acceptance rate on identical briefs
Precision of edits and unchanged-region preservation
Total spend through the final accepted image
Run a fair comparison
Use the same representative inputs and a written answer rubric. Pin the model version and record the endpoint, thinking configuration, quantization when known, context length, output allowance and concurrency. Preserve native request differences in an adapter rather than silently dropping fields.
Measure full request time, error rate and reported usage. For tools, validate arguments and score the result after execution. For images, preserve resolution and preprocessing. A valid response format alone does not establish a correct answer.
Compare cost per accepted result
Calculate input cost plus all billed output, including reasoning where the provider counts it. Add retries and discarded answers. Divide the total by the number of results that pass your quality criteria. Keep API usage cost separate from deployment and engineering cost.
Artificial Analysis observations are refreshed by the integrated Radar. Differences in artifact, benchmark version or method prevent a score comparison. Read the evidence methodology →
YOUR WORKLOAD
Compare cost per accepted task.
Enter results from your own evaluation. A task should count once it meets your written quality criteria. Include failed attempts and retries in total spend. These inputs stay in this browser and are not a Flonno benchmark.
GPT Image 2.5 Flare
Enter all four values. Attempted tasks must be at least one; accepted tasks cannot exceed attempted tasks.
GPT Image 2.5 Sunburst
Enter all four values. Attempted tasks must be at least one; accepted tasks cannot exceed attempted tasks.
Turn evidence into a deployment decision
A source benchmark can inform your shortlist, while your evaluation shows whether a model finishes the work at an acceptable cost. Keep the chosen endpoint, processing terms and output limits in the decision record. Inclusion in the Radar does not mean the model is served by Flonno.
