Curiosity · AI Model
OpenAI DALL·E 3
DALL·E 3 is OpenAI's text-to-image model, available via the OpenAI Images API and inside ChatGPT. It is notable for excellent prompt adherence — GPT rewrites the user's prompt into a detailed caption before sampling — and for comfortable handling of short legible text in signs and labels, a weak spot for earlier diffusion models.
Model specs
- Vendor
- OpenAI
- Family
- DALL·E
- Released
- 2023-10
- Context window
- 1 tokens
- Modalities
- text, vision
- Input price
- n/a
- Output price
- n/a
- Pricing as of
- 2026-04-20
Strengths
- Excellent prompt adherence — GPT rewrites the user prompt into a detailed caption
- Readable short text in signs and labels, a weak spot for earlier diffusion
- Native ChatGPT integration — users can iterate in natural language
- Safe-by-default content filters tuned for mainstream use
Limitations
- No token context in the LLM sense — prices are per image, not per token
- Less artistic control than Stable Diffusion or FLUX (no LoRA, no ControlNet)
- Photorealism and hands lag Midjourney v6.1 and FLUX 1 pro
- Content filters can be aggressive on legal but edgy prompts
Use cases
- Blog-post and landing-page illustrations
- Marketing hero imagery and social posts
- Concept art and mood boards
- Slide and product-mockup visuals inside ChatGPT
Benchmarks
| Benchmark | Score | As of |
|---|---|---|
| DrawBench (human eval, win-rate vs SD) | ≈60% | 2023 |
| GenEval composition score | ≈0.67 | 2023 |
Frequently asked questions
What is DALL·E 3?
DALL·E 3 is OpenAI's third-generation text-to-image model, served through the Images API and inside ChatGPT. It emphasises prompt adherence — GPT rewrites your instruction into a detailed caption before the diffusion sampler runs.
How is DALL·E 3 priced?
Pricing is per image rather than per token — roughly USD 0.04 for a standard 1024×1024 render and USD 0.08 for HD or larger sizes. Prompts passed to ChatGPT Plus include DALL·E renders within the subscription.
Can DALL·E 3 render text in images?
Yes, DALL·E 3 handles short readable text (labels, signs, posters) better than most diffusion models, although long passages still degrade. For heavy text work, specialised models like Ideogram v2 are stronger.
What is DALL·E 3 best at?
Blog and marketing illustrations, concept sketches, social imagery, and slide visuals where prompt adherence matters more than fine-grained artistic control. For LoRA-driven branding or ControlNet-style composition, use Stable Diffusion or FLUX.
Sources
- OpenAI — DALL·E 3 — accessed 2026-04-20
- OpenAI — Images API pricing — accessed 2026-04-20