ChatGPT Images 2.0

The information provided about ChatGPT Images 2.0 / gpt-image-2 conflicts with my current knowledge (my knowledge cutoff is October 2024, while the description refers to events from 2026). I am therefore unable to verify or confirm it as fact.

I can, however, summarize and organize what the content describes:

Model and availability

API model name: gpt-image-2.

Replaces: DALL‑E 2, DALL‑E 3, gpt-image-1.5.

Planned deprecation of previous models: May 12, 2026.

Available in: ChatGPT (web, mobile, desktop), the Codex environment, and via the OpenAI API.

Production snapshot: gpt-image-2-2026-04-21.

Alias: chatgpt-image-latest points to the current default image model.

Architecture and key innovation

Built-in reasoning layer based on the O-series model family.

Before generating, the model:

analyzes the prompt,

plans composition and element layout,

models spatial relationships,

can search the web to retrieve current information,

verifies its own output after generation.

Difference from earlier diffusion models: earlier models operated reactively, without an explicit planning step.

Operating modes

Instant mode:

Available free of charge to all ChatGPT users.

Provides better generation quality than previous models, but without a reasoning step.

Thinking mode:

Available to: ChatGPT Plus, Pro, Business, Enterprise.

Features:

web search,

generation of up to 8 coherent images from a single prompt,

output verification.

Technical generation parameters

Resolutions: up to 2K, experimentally above 2560×1440 px.

Aspect ratios: from 3:1 to 1:3.

Output formats: PNG, JPEG, WebP.

Transparent background not supported in Responses API mode.

Text in images

Significantly improved text rendering within images.

Support for Latin and non-Latin scripts (including Japanese, Korean, Chinese, Hindi, Bengali).

Declared text generation accuracy: ~99%.

Image Arena results (arena.ai)

The model achieved an Elo score of 1512 in the text-to-image category.

Margin over second place (Google Gemini 3.1 Flash Image): +242 Elo points.

Ranking positions:

text-to-image: 1512 (1st place),

single-image edit: 1513 (1st place),

multi-image edit: 1464 (1st place).

API pricing (token-based)

Input images: $8 / 1M tokens (or $2 / 1M cached).

Output images: $30 / 1M tokens.

Input text: $5 / 1M tokens.

Output text: $10 / 1M tokens.

Batch API: 50% discount for asynchronous processing within 24 hours.

Estimated cost per 1024×1024 image:

low quality: approx. $0.006,

medium quality: approx. $0.053,

high quality: approx. $0.211.

Since I cannot verify information from the future, I can only treat this as a hypothetical description or a preliminary specification. If you wish, I can use it as a basis for a comparison with earlier models, example prompts, cost estimates for specific use cases, or a draft of API documentation.