AI image models: GPT Image 2 leads new artificial analysis categories

The artificial analysis framework has broadened its text-to-image arena, adding two use scenarios—UI/UX design and live-action film—alongside two measurement capabilities: physics and text rendering. The shift moves evaluation from a single overall ranking to task-specific segmentation, positioning OpenAI’s GPT Image 2 at the top of three of the four new categories.

In the UI/UX category, which covers mockups for applications, vehicles, spatial environments, dashboards, icons and design-system assets, GPT Image 2 and Reve 2.1 share first place. Nano Banana 2 ranks third and is marked as best value, indicating a more favorable cost-performance ratio for users willing to sacrifice the absolute top spot.

For the live-action film scenario—photorealistic cinematic scenes, storyboards and character sheets—GPT Image 2 leads by a margin, followed by Nano Banana Pro, with Reve 2.1 in third place. Here there is no tie: the OpenAI model demonstrates clear superiority in producing frames that resemble real photography.

In the physics test (gravity, support, collision, state changes and temperature), GPT Image 2 is first, GPT Image 1.5 second and Nano Banana 2 third and cheapest of all. In text rendering—long text, small text, symbols and artistic letters—GPT Image 2 again ranks first, Reve 2.1 second, while Nano Banana 2 offers the best value at $67 per thousand images, versus $211 and $200 for the two competitors. The quality gap here is the widest measured: 178 Elo points separate the strongest and weakest model among the ten evaluated.

The complete methodology, including instruction authoring, refresh cycles, retirement criteria and scoring formula, is published on the website. The new approach reflects the view that a single overall ranking is no longer sufficient: developers and designers need to know which model fits their specific task and its commercial-scale cost.