Frontier image models

last edited

Frontier image models are the systems at the top of blind human-preference boards. In November 2026 the leaders are closed models served through an assistant or API. OpenAI holds first place. The models reference hub covers the full landscape, and other striking models sit one tier down.

Who holds the frontier

On the Artificial Analysis text-to-image board, GPT Image 2.6 ranks first at about 1261. The board covers 110 current models. Votes become an Elo rating, where a 100-point gap means about 64% of votes go to the higher model.

Rivals sit roughly 85 to 95 points behind. Nano Banana 2.1 scores 1176. MAI-Image-2.6 scores 1170. The xAI image model scores 1168. Meta's Muse Image scores 1128.

Recent releases

OpenAI released Images 2.6 in November 2026. It added GIF output, saved references for consistent characters and faster generation. Flare can produce images in as little as six seconds, per the OpenAI announcement.

Google made Nano Banana 2.1 generally available in October 2026. The older Nano Banana 2 is deprecated, and the Gemini API changelog tells developers to migrate. Its API price ranges from $0.034 to $0.134 per image.

Below the frontier

ByteDance's latest model scores 1135. Its price is $0.020 to $0.090 per image. The flagship FLUX 3 Image scores 1115. Prices run from $0.003 to $0.080. Qwen-Image-3.0-Pro scores 1096.

Fibo 2.0 scores 1037. A second-tier model from another image lab scores 1035. See FLUX in production, FLUX LoRA training and open models. Locally runnable families rank lower but offer the cheapest pricing.

Midjourney and deployment

Midjourney has no public API, so it appears on no API-based board. Frontier labs bill through APIs. The FLUX developer offers an API plus open weights.

Prices are per 1024px image, using creator API pricing listed by the arena. See enterprise deployments, the workflows hub, platforms and APIs and platform migration.

Sources