GPT Image 2.0 (ex-DALL-E)
Overview
GPT Image 2 is OpenAI's image generation model, powering both image generation inside ChatGPT (ChatGPT Images 2.0) and the OpenAI API. It shipped on April 21, 2026, and ChatGPT image generation switched to it across every plan including the free tier. In the API it is called as gpt-image-2.
Key Features
Native reasoning before generation
GPT Image 2 is OpenAI's first image model with reasoning built into the architecture — it plans the layout, can search the web for references, and self-checks the output before rendering. That pays off on vague briefs and dense, information-heavy graphics.
Text rendering and multi-image consistency
Character-level text accuracy reaches roughly 99% across Latin, CJK, Hindi and Bengali scripts, and Thinking mode can produce up to eight images from one prompt with consistent characters, objects and style. It supports resolutions up to 2K and aspect ratios from 3:1 to 1:3.
Conversational editing inside ChatGPT
The practical advantage is iterating in the chat itself: hand it a reference image and refine with plain language. The same model is available through the API as gpt-image-2.
Plans & Pricing
| Entry point | Price | Notes |
|---|---|---|
| ChatGPT Free | $0 | A few images per day |
| ChatGPT Plus | $20/mo | Higher daily cap |
| ChatGPT Pro | $200/mo | Near-unlimited |
| API (gpt-image-2) | Token-based | $30 per 1M image output tokens, etc. |
The API bills on tokens like the text models: $8 per 1M image input tokens, $2 cached image input, $30 image output and $5 text input. At 1024×1024 that works out to roughly $0.006 low quality, $0.053 medium and $0.211 high. The Batch API halves all of these.
Typical Use Cases
- Banners, thumbnails and diagrams that contain text
- Illustrations for decks and documentation
- Variations and localized edits from a reference image
- High-volume generation via the API (50% cheaper on Batch)
Strengths
- Available directly in ChatGPT with conversational iteration
- Multilingual text rendering is genuinely production-grade
- Strong prompt adherence on complex compositions
- One model ID covers everything, making API integration simple
Weaknesses
- Thinking mode makes generation noticeably slower
- Less stylistically distinctive than Midjourney, less photoreal than Flux
- High-quality settings get expensive per image
Official Info
- ChatGPT Images 2.0: https://openai.com/index/introducing-chatgpt-images-2-0/
- API docs: https://platform.openai.com/docs/models