You cannot choose DALL-E 3 any more. OpenAI’s deprecations page lists both dall-e-2 and dall-e-3 as removed from its API on 12 May 2026, after an announcement on 14 Nov 2025, and points developers to gpt-image-2, gpt-image-1 or gpt-image-1-mini. So the real comparison is Midjourney against OpenAI’s GPT Image models. Independent write-ups agree on the split: GPT Image leads on legible text and following detailed instructions, Midjourney on painterly, art-directed looks.
Retirement log
| Date | What happened | Source |
|---|---|---|
| 14 Nov 2025 | OpenAI announces removal of dall-e-2 and dall-e-3 from the API | OpenAI deprecations page |
| Dec 2025 | ChatGPT’s image generation reported to be on GPT Image 1.5 rather than DALL-E | Third-party reports (for example Genra) |
| 21 Apr 2026 | Current default gpt-image-2 snapshot, gpt-image-2-2026-04-21 | OpenAI model page |
| 12 May 2026 | dall-e-2 and dall-e-3 removed from the API | OpenAI deprecations page |
Anything you read that recommends DALL-E 3 for text-in-image work, or lists its $0.04 to $0.12 API prices, describes a product that no longer exists. This page’s earlier version did.
Which to pick
| If you mostly make | Lean towards | Why (from published comparisons) |
|---|---|---|
| Thumbnails, posters and graphics with words in them | GPT Image | Third-party comparisons rate its text and layout accuracy well above Midjourney’s |
| Edits to an existing image (“change the background”) | GPT Image | It supports an edits endpoint; the model page lists generation and editing |
| Mood, concept art, fashion or editorial looks | Midjourney | Comparisons keep giving it the edge on aesthetic quality |
| A consistent visual style across a series | Midjourney | Style and personalisation controls are its strength, per the same write-ups |
| Occasional use with a ChatGPT subscription already paid | GPT Image through ChatGPT | No second subscription |
| Programmatic generation in your own app | GPT Image API | Midjourney is not an API product in the same sense; check its current terms |
These are the claims of other publishers, not measurements of ours. Vendor-run comparison sites and blogs have an interest in one side, so weigh the ones that show the same prompts and outputs side by side.
Prices, now readable
- Midjourney: Basic $10, Standard $30, Pro $60, Mega $120 a month, sold in fast GPU hours, with unlimited Relax mode from Standard; extra fast time is $4 an hour (plan page read 29 Sep 2026). Full table in the Midjourney review.
- OpenAI’s image models: billed in tokens on the API, not per image. The API pricing page lists gpt-image-2.5-sunburst and gpt-image-2.5-flare at $8.00 per million image input tokens and $30.00 per million image output tokens (text input $5.00), and points to a calculator for per-image estimates. The gpt-image-2 model page (snapshot 21 Apr 2026) gives no per-image figures, so none are quoted here. ChatGPT plan prices were not readable on the date checked.
- Terms: OpenAI’s Terms of Use say that, as between you and OpenAI, you own the output. Midjourney says you own what you create, with a $1,000,000 revenue rule. Both compared with Ideogram, Leonardo and Adobe in best AI image generators.
What we could not verify
- Any quality score. We did not generate images; treat “better” as each source’s opinion.
- A per-image price for GPT Image. Token billing means the cost per image depends on size and quality settings; use OpenAI’s calculator.
Method
Researched from OpenAI’s API documentation and third-party comparisons; checked 29 Sep 2026. No images were generated for this page.
Related
Other image and video tools are in the vault; for text-in-image specialists see Ideogram, and for another retired product see Sora alternatives.