DALL-E 3 shuts down in 20 days — here's the migration plan

Updated May 11, 2026: DALL-E 3 shuts down tomorrow, May 12, 2026. OpenAI is sunsetting both DALL-E 2 and DALL-E 3. The deprecation notice went out to developers using DALL-E model snapshots on November 14, 2025. Since December 2025, DALL-E 3 has already been pulled from the ChatGPT consumer interface without warning.

If you’ve built anything on the dall-e-3 or dall-e-2 model IDs in the OpenAI API and haven’t migrated yet, the cutover happens in the next ~24 hours. This is the last call.

The replacement path

OpenAI’s official migration guidance points users to:

• gpt-image-1.5 — the current flagship image model, integrated into GPT-5.4’s multimodal architecture
• gpt-image-1-mini — cheaper, faster variant for high-volume applications
• GPT Image 2 — rumored to launch between late April and mid-May 2026, likely before or shortly after the May 12 cutover. Analysts expect it alongside a GPT-5.4 update.

API pricing comparison:

The token-based pricing on gpt-image-1.5 is more variable than DALL-E 3’s flat per-image model. For most straightforward use cases, gpt-image-1.5 at standard quality is similar in cost to DALL-E 3 HD. For high-quality generation, it can be 2-3x more. For batch high-volume work, gpt-image-1-mini is actually meaningfully cheaper than DALL-E 3 ever was.

Migration steps for developers

1. Audit your DALL-E 3 API calls. Grep your codebase for "dall-e-3" and "dall-e-2" strings.
2. Swap the model ID. In most OpenAI SDKs this is a one-line change — model: "dall-e-3" → model: "gpt-image-1.5".
3. Test prompt behavior. GPT Image 1.5 follows prompts more literally than DALL-E 3 did. A prompt that was vague but produced good DALL-E 3 output may need tightening for GPT Image 1.5.
4. Check text rendering. GPT Image 1.5 is dramatically better at rendering text in images than DALL-E 3 was. If you previously post-processed images to add text, you may not need to anymore.
5. Update pricing math. Token-based means the bill depends on quality and resolution. Run a week of production traffic through GPT Image 1.5 at standard quality before committing to a tier.

If you’re on a framework like LangChain, LlamaIndex, or Vercel AI SDK: check their migration docs — most have already added GPT Image 1.5 support.

If you can’t migrate by May 12: consider pausing that feature and restoring it after May 12 on GPT Image 1.5 rather than keeping a DALL-E 3 code path that won’t work. Partial migration is worse than a clean cutover on a hard deadline.

The consumer-facing story

If you only use DALL-E via the ChatGPT web interface, you’ve likely already been routed to GPT Image 1.5 without noticing. OpenAI switched consumer traffic over in December. The May 12 date is strictly the API sunset.

If you’re paying for ChatGPT Plus ($20/mo): You already get GPT Image 1.5 plus the new ChatGPT Images 2.0 “thinking” features that shipped April 21, 2026. No action required.

Alternatives outside OpenAI

Worth comparing before migrating deep into GPT Image 1.5:

• Nano Banana Pro (Google Gemini) — Fastest mainstream image model (1-3 sec per image), strong photorealism, free tier genuinely usable, API cost competitive with or below GPT Image
• Midjourney — Artistic quality leader, but no public API (web/Discord only), so not a straight DALL-E 3 API replacement
• FLUX.2 via Black Forest Labs — Open-weight option for self-hosting, strongest quality among open models in 2026
• Adobe Firefly — IP-safe commercial option with contractual coverage, API available through Firefly Services

For pure API replacement, my take:

• Default: gpt-image-1.5 (minimal code change from DALL-E 3)
• Cost-sensitive: Nano Banana Pro API (cheaper, faster, comparable quality)
• Commercial/brand-safe: Adobe Firefly
• Maximum quality control: self-host FLUX.2

The deeper question

DALL-E 3’s sunset, Sora’s April 26 shutdown, and the quiet promotion of GPT Image and Codex all point the same direction: OpenAI is consolidating its lineup around GPT-5.x-integrated multimodal models. Separate product brands (DALL-E, Sora) are getting absorbed into or replaced by GPT Image and GPT Video successors.

If you’ve been relying on a single-purpose OpenAI product, the pattern should make you plan for it to be sunsetted eventually. DALL-E 3 was the most-used AI image generator in the world three months ago. By May 12, 2026, it’s gone.

Migrate early. Don’t wait until May 10.

Related:

• Full DALL-E / OpenAI Image review
• Best AI image generation tools in 2026
• Sora shuts down in 3 days

Questions or corrections? Email Pick Right. Want the full list? See all news.