GPT Image 2 Launch Timeline: What We Actually Know in April 2026
OpenAI has stayed quiet, but three codename models on LM Arena, an internal ChatGPT A/B, and the May 12 DALL-E retirement deadline make the launch window easy to bracket.
You cannot call GPT Image 2 by name on any hosted provider, including fal.ai. That is not the same as not having information. OpenAI quietly pushed three codename models (packingtape-alpha, maskingtape-alpha, gaffertape-alpha) to LM Arena on April 4, 2026. They were pulled after about 36 hours. That is the strongest public signal the model exists and is close enough to production to be benchmarked against the field.
The DALL-E retirement deadline
OpenAI announced that DALL-E 2 and DALL-E 3 retire on May 12, 2026. OpenAI does not retire a product line without a replacement in the API, which pins the upper bound on GPT Image 2's public availability to a week or two before that date. If the replacement slipped past May 12, the API surface would have a hole, and OpenAI has not done that with any other family.
The ChatGPT A/B
Power users on Twitter have posted renders from ChatGPT that do not match any public model. The text rendering is noticeably cleaner, the yellow cast from 1.5 is gone, and the file metadata hints at a new model string. That is consistent with an internal A/B rollout, which usually precedes a public ship by two to six weeks.
What to do with that
Ship your switch now. Add a feature flag that picks fal-ai/gpt-image-2 when it lights up and falls back to fal-ai/gpt-image-1.5 otherwise. Subscribe to the fal.ai changelog RSS, set a Slack reminder for May 5 and again for May 12. When the endpoint lights up, your production traffic flips with a deploy.
The rest is waiting.