Now superseded in ChatGPT by Images 2.5, so this matters mainly to developers still calling gpt-image-2 in the API.
Kenning's take
What it does
ChatGPT Images 2.0 is the image generation model OpenAI released in April 2026, sold in the API as gpt-image-2. OpenAI says it improved world knowledge, instruction following and dense text, and added a thinking mode that can search the web and make several images from one prompt. It has since been followed by Images 2.5 in ChatGPT, while gpt-image-2 remains listed in the API.
- Better rendering of dense text
- Improved world knowledge and instruction following
- Thinking mode with reasoning and live web search
- Several images from a single prompt
- Image generation and image edit endpoints in the API
Who it suits. Developers using OpenAI's Images API and people comparing image model versions.
Key facts
- Category
- AI models
- Part of
- ChatGPT
- Made by
- OpenAI
- Based in
- United States
- Released
- 21 April 2026
- How to get it
- API model gpt-image-2 (image generation and edit endpoints); in ChatGPT it has been succeeded by Images 2.5
- Works on
- API
- In the UK
- Yes
- API price
- $5 per million text input tokens, $8 per million image input tokens, $30 per million image output tokens (standard tier)
FAQ
When was ChatGPT Images 2.0 released?
OpenAI published the Images 2.0 system card and launch on 21 April 2026.
What is the API name for Images 2.0?
gpt-image-2, with snapshot gpt-image-2-2026-04-21.
How much does gpt-image-2 cost?
$5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
Astra for Law
OpenAI's GPT-6 Astra set up with a US legal search index for legal research and drafting
ChatGPT Images 2.5
OpenAI's September 2026 image model with sharper detail, steadier edits and up to 50% faster generation
GPT-5.1-Codex-Max
OpenAI coding model built for long agent sessions, now marked deprecated in the API
GPT-5.4
OpenAI model from March 2026 with built-in computer use and a 1M-token context