OpenAI's image generation, formerly DALL·E, now runs as ChatGPT Images inside ChatGPT and as GPT Image models in the OpenAI API.
✓ Pros
- Images can be created and revised by conversation in ChatGPT, including on the Free plan within limits
- The same family of image models is available through the API with published per-token prices
- Sketch input and templates in ChatGPT let users start from a rough drawing or a format such as a poster
- Outputs carry C2PA metadata and invisible watermarking, which helps where provenance must be shown
✗ Cons
- API models are retired on a schedule (gpt-image-1 ends on 23 October 2026), forcing migrations
- Cloud only; there are no weights to run on your own infrastructure
- Token billing makes per-image cost depend on size and quality settings, so estimates need OpenAI's calculator
- OpenAI's prompt and image safety checks may refuse some creative requests
| Free tier | ✓ Free tier |
| Pricing model | subscription+usage |
| Price (ChatGPT Plus) | $20 USD |
| Features | |
| API | ✓ Available Docs ↗ |
| Pricing Plans | Free$0ChatGPT無料プランで回数制限付きの画像生成 ChatGPT Plus$20/月チャット内で画像生成 API（GPT Image）トークン課金画像サイズと品質で費用が変わる。旧DALL-E 2/3は2026年5月にAPIから削除 |
| Platforms | |
| Integrations | ChatGPT, OpenAI API, Microsoft Copilot, Bing Image Creator, Azure OpenAI |
| Homepage | https://chatgpt.com/images/ |
Overview
This entry covers OpenAI's image generation, which has replaced DALL·E: the DALL·E models were removed from the API in May 2026. In ChatGPT, images are made and revised through conversation, and every tier including Free can generate them, with higher limits on paid plans. Developers call GPT Image models through the API, billed per token for text input, image input and image output; OpenAI recommends GPT-Image-2.5 Flare as the default for most applications. Older GPT Image models are retired on a schedule, so integrations need periodic migration. There is no self-hosted option.
