OpenAI's ChatGPT Images 2.0 adds web search and multi-image generation with reasoning mode

OpenAI released ChatGPT Images 2.0, powered by the new GPT Image 2 model. The update enables web search integration for paid subscribers in thinking mode, generates up to eight images from a single prompt while maintaining visual consistency, and supports 2K resolution output.

Key capabilities

The thinking mode feature is available exclusively to ChatGPT Plus, Pro, Business, and Enterprise subscribers. According to OpenAI, when thinking is enabled, the image generator can search the web for reference information, analyze uploaded files to create visual explainers, and "reason through the structure of the image before generating."

ChatGPT Images 2.0 can produce up to eight images simultaneously in thinking mode while maintaining consistent characters, objects, and styles across scenes. OpenAI says this functionality is designed for use cases including manga page sequences, social media graphic series, and multi-room design plans.

Technical improvements

All ChatGPT users gain access to enhanced image style capabilities, including improved rendering of photos, pixel art, manga, and cinematic stills. The system now outputs images at up to 2K resolution and supports aspect ratios from 3:1 (wide) to 1:3 (tall).

Text generation accuracy improved significantly across non-Latin scripts. OpenAI claims "significant gains" for Japanese, Korean, Chinese, Hindi, and Bengali text rendering, expanding beyond its previous focus on English and Latin-based languages.

Release context

OpenAI first launched ChatGPT Images in 2025, with a previous major update in December 2025 that added faster generation speeds and photo editing tools. The company positions this release against competing systems including Google's Nano Banana Pro and Microsoft's MAI-Image-2.

Pricing for paid tiers remains unchanged. ChatGPT Images 2.0 rolled out to all ChatGPT and Codex users on April 21, 2026.

What this means

The web search integration represents a significant architectural shift, allowing the image model to reference real-world information rather than relying solely on training data. The multi-image generation with style consistency addresses a practical limitation in sequential visual content creation. However, the restriction of thinking mode to paid tiers creates a clear capability divide between free and premium users, particularly for professional use cases requiring batch generation or research-backed imagery.