OpenAI Images

OpenAI image generation — gpt-image-2 (flagship, reasoning-based), gpt-image-1 (native multimodal), dall-e-3 (legacy), dall-e-2 (legacy).

GPT Image 2

Flagship OpenAI image model (Apr 2026) — reasoning-based composition, near-perfect multilingual text rendering (CJK, Arabic, Devanagari), supports up to 3840px on the long edge, accepts up to 16 input images for composite edits.

2 methods

GPT Image 1

Native GPT-4o image model — t2i + image edit, low/medium/high quality.

2 methods

DALL·E 3

Legacy DALL·E 3 — text-to-image only, standard or HD.

1 method

DALL·E 2

Legacy DALL·E 2 — t2i + edit at 256/512/1024 squares.

2 methods

GPT-4o mini (vision)

Budget multimodal model for image understanding. NB: inflates image tokens ~33x — gpt-4.1-mini is cheaper and better for vision.

1 method

GPT-4.1 mini (vision)

Cheap, reliable multimodal model — image → text or strict JSON. Recommended OpenAI vision option.

1 method

GPT-5 mini (vision)

Lowest-cost OpenAI vision model (reasoning family) — image → text or strict JSON.

1 method

GPT-4.1 (vision)

High-accuracy multimodal model — image → text or strict JSON. For accuracy-sensitive jobs.

1 method

GPT-4o (vision)

Flagship multimodal model — image → text or strict JSON. Highest accuracy, highest cost.

1 method