OpenAI has deprecated the gpt-image-2 model alias and the pinned gpt-image-2-2026-04-21 snapshot. Both are scheduled to stop serving API requests on February 10, 2027.
What changed
Applications that explicitly request either model now have a fixed migration deadline. OpenAI recommends gpt-image-2.6-sunburst or gpt-image-2.6-flare as replacements. The deprecation is effective immediately; shutdown means the retired identifiers will no longer be accessible after the stated date.
The practical consequence is compatibility risk rather than an immediate outage. Teams should inventory model names in source code, environment variables, prompt configurations, evaluation fixtures, and fallback routes, then test the replacement model before changing production traffic. Code that relies on the pinned April snapshot is affected as well as code using the moving alias.
Migration requires output testing
Changing an image model can alter instruction following, editing behavior, style, text rendering, latency, safety filtering, output dimensions, and cost. A successful API call is therefore not enough evidence of a safe migration. Existing applications should rerun representative generation and edit cases, compare rejected-request behavior, verify transparent-background and file-format requirements, and confirm that downstream image processing still accepts the results.
OpenAI’s current image documentation lists both recommended 2.6 variants, but the deprecation notice does not provide comparative price, latency, or quality benchmarks for this migration. DrComps has not independently benchmarked the replacement models, so the recommendation should not be read as proof that either replacement is universally faster, cheaper, or better for every workload.
What is not changing today
The notice does not say that gpt-image-2 has stopped working now, nor does it announce a change to the Images API itself. The risk is deferred but concrete: integrations left on either retired identifier are expected to fail after February 10, 2027.