If you got an email from OpenAI this week, the subject line was easy to overlook. DALL-E 2 and DALL-E 3 are being retired on May 12, 2026. The recommended replacements are gpt-image-1.5 and gpt-image-1-mini. OpenAI is framing this as an upgrade — and for some workloads it is. For others, the math doesn't work the same way.
What Changed
Both DALL-E models are being sunset. After May 12, API calls to either model will stop working. The migration path is to gpt-image-1.5 (the current flagship) or gpt-image-1-mini (the budget option).
The pricing comparison is where it gets complicated. DALL-E 3 standard was $0.04 per image, HD was $0.12. gpt-image-1-mini at low quality starts at $0.005 — that's up to 87% cheaper for high-volume, lower-quality workloads. But gpt-image-1.5 at high quality reaches $0.20 per image, nearly double DALL-E 3 HD. And the new models add token-based charges on top of the per-image price — text input tokens, image input tokens for edits, output tokens. The per-image number in the docs is not the whole invoice.
The Bigger Picture
This is part of a broader pattern: OpenAI retiring older model families as the GPT Image series matures. The framing is always "better model, simpler pricing" — but simpler doesn't always mean cheaper, especially when token-based overhead is introduced alongside the per-image rate. The same shift happened with GPT-4 → GPT-4o transitions, where the headline price dropped but usage patterns meant real-world bills varied significantly.
What This Means for You
34 days. If you have DALL-E 2 or DALL-E 3 hardcoded in any pipeline, it breaks on May 12. The migration itself is a model string change — straightforward. The part that requires attention is auditing your actual cost exposure under the new pricing model before you switch. High-quality image generation at scale could cost significantly more. Run the numbers for your specific usage pattern before assuming the new model is a direct cost-equivalent replacement.
Trish @ StackDrift
StackDrift monitors OpenAI API and pricing pages. When the May 12 retirement date approaches, we'll flag any changes to the migration timeline. Subscribe here.