← Back to Blog

OpenAI Retires DALL·E 2 and DALL·E 3 Today, Pushing Image API Teams Into the GPT Image Era

Editorial image for OpenAI Retires DALL·E 2 and DALL·E 3 Today, Pushing Image API Teams Into the GPT Image Era about Developer Tools.

Key Takeaways

  • OpenAI’s API shutdown for DALL·E 2 and DALL·E 3 takes effect on May 12, 2026.
  • OpenAI’s published replacement guidance points developers to gpt-image-1 or gpt-image-1-mini.
  • The bigger shift is OpenAI moving image generation into its broader GPT Image stack, not just retiring one brand.
  • Teams with production image workflows now need to revalidate outputs, tooling, and workflow assumptions immediately.
BLOOMIE
POWERED BY NEROVA

On May 12, 2026, OpenAI’s API retirement for DALL·E 2 and DALL·E 3 takes effect, ending access to two of the company’s older image-generation models and forcing developers onto newer image paths. OpenAI had notified developers on November 14, 2025 that both models would be removed from the API on this date, with gpt-image-1 or gpt-image-1-mini listed as the recommended replacements.

That makes today’s change more than a housekeeping update. Any business still relying on dall-e-2 or dall-e-3 inside production content, design, or asset-generation workflows now has a live dependency problem to solve.

What ends on May 12

OpenAI’s deprecations documentation puts both DALL·E models on the same shutdown date: May 12, 2026. The company’s guidance is explicit that API users should move off those model names and onto the GPT Image path instead.

OpenAI image-model shutdowns taking effect on May 12, 2026

ModelStatus on May 12, 2026OpenAI replacement guidance
DALL·E 2Removed from the APIgpt-image-1 or gpt-image-1-mini
DALL·E 3Removed from the APIgpt-image-1 or gpt-image-1-mini

OpenAI’s DALL·E 3 help documentation now also points developers toward the newer GPT Image API, reinforcing that the company no longer sees DALL·E as the forward product surface for image generation in the API.

Why this is bigger than one model retirement

The important shift is not only that DALL·E is going away. It is that OpenAI is consolidating image generation around a broader multimodal stack rather than a standalone legacy image brand.

In newer help-center materials, OpenAI describes the GPT Image API as the current path for creating, editing, and transforming images with its latest multimodal image models. That reframes image generation as part of a larger model family, not a separate product lane.

For buyers and builders, that changes the operating assumption. Image generation is becoming another fast-moving capability inside a general AI platform, which means teams can no longer treat the model layer as stable for years at a time.

Where businesses will feel the disruption first

The most exposed teams are the ones that embedded DALL·E calls deep inside repeatable business processes rather than occasional experiments. That includes:

  • Marketing ops generating campaign creative variations automatically
  • Ecommerce teams producing product visuals, background edits, or merchandising assets
  • Internal design systems that route prompts through older OpenAI image endpoints
  • Agent workflows that create visual assets as one step inside a larger automation chain

For those teams, the problem is not just swapping one model name for another. Output style, edit behavior, moderation patterns, latency, and cost assumptions can all shift when the underlying image system changes. A workflow that looked stable in April can break in May even if the business logic around it stayed the same.

What this signals for AI agents and automation

The practical lesson is broader than image generation. AI systems that matter to a business need to be built so they can survive model churn, not just exploit the latest release.

If a workflow depends on one exact model alias, one vendor-specific behavior, or one legacy endpoint, it is fragile by default. The stronger pattern is to build around the business outcome: generate an approved asset, edit a product image, produce a campaign draft, or package a visual for review. The model can change underneath that outcome layer, but the workflow should keep running.

That is why today’s DALL·E shutdown matters. It is a small API event on the surface, but it is also a reminder that enterprise AI, automation, and agent systems need portability, fallback logic, and regular revalidation built in from the start.

Build workflows that survive model churn

If a model retirement can disrupt part of your stack, the next step is designing around outcomes instead of one fragile endpoint. Explore Nerova’s agents and AI teams to see how businesses package content, support, and operations into more resilient workflows.

Browse AI agents and teams
Ask Bloomie about this article