GPT-Image-2: The New #1 on the Arena Leaderboard

OpenAI's GPT-Image-2 (also branded ChatGPT Images 2.0) launched in late April 2026 and immediately took the #1 spot on the Arena text-to-image leaderboard. With an Elo of 1,512, it leads second-place Google Nano Banana 2 by 241 points — the largest gap between #1 and #2 ever recorded on Arena. For perspective, positions 4 through 15 on the same leaderboard span only 92 points combined.

GPT-Image-2 is a fully standalone image model, no longer tied to the GPT-4o pipeline. It introduces a native reasoning ("thinking") mode that can search the web for context and generate up to 8 consistent images from a single prompt, and renders text at roughly 99% accuracy — a clear step up from the 90–95% typical of previous-generation models. The model supports up to 4K output and flexible aspect ratios.

On the API, GPT-Image-2 accepts a text prompt plus optional reference images to combine subjects, styles, or scenes into a single output. You can pick low / medium / high / auto quality, choose 1:1, 3:2, or 2:3 aspect ratios, set the background to transparent or opaque, and export in PNG, JPEG, or WebP.

The combination of pixel-level realism, near-perfect text rendering, and reasoning makes GPT-Image-2 especially strong for posters, infographics, packaging, app mockups, e-commerce visuals, and any creative work where typography matters.

Use Case 1: Product Packaging Mockups

Use Case 2: UI and App Screen Mockups

We've added GPT-Image-2 to Hocha's text-to-image generator with all four quality tiers (low / medium / high / auto), reference image support, and 1:1 / 3:2 / 2:3 aspect ratios. You can give it a try right now.