Introducing ChatGPT Images 2.0

The framing here is that text-in-image accuracy was a known problem and this fixes it, but OpenAI hasn't released independent benchmark comparisons against its prior model or against competitors — so 'improved' is self-reported. Multilingual support sounds significant, but the languages covered and the quality floor for non-Latin scripts remain unspecified.

Image generation is suddenly a crowded front. Just days before this announcement, Google extended Gemini's image capabilities by connecting it directly to users' Google Photos libraries (covered here April 16 via The Verge and Ars Technica), which is a meaningfully different competitive angle: personalization over raw rendering quality. OpenAI is iterating on output fidelity while Google is betting on personal context as the differentiator. Neither approach is obviously correct, and the two strategies may appeal to different user segments entirely. What's missing from OpenAI's announcement is any acknowledgment of that personalization gap.

Watch whether independent image-quality evaluations, particularly on non-Latin script rendering in languages like Arabic or Hindi, confirm the multilingual claims within the next 60 days. If third-party testers find meaningful regressions or parity with the prior model, the 'addresses longstanding weaknesses' framing collapses.