GPT Image 2 vs GPT Image 1.5 — which should I use?

GPT Image 2 is the reasoning, quality-first default and wins on prompt adherence and text. GPT Image 1.5 (Dec 16, 2025) remains available as a stable legacy option and is better if you need transparent backgrounds or input-fidelity control, which GPT Image 2 dropped.

Does GPT Image 2 output transparent backgrounds?

Not on its text-to-image endpoint — requests with background: transparent aren't supported. For transparent PNGs, product cut-outs, logos or UI assets, use gpt-image-1.5.

Is 'GPT Image 2' the official name?

Yes. The official API model name is gpt-image-2, released April 21, 2026. OpenAI has not used 'DALL·E 4' — that is not an official name.

GPT Image 2: specs, access and what it's best at

Updated July 22, 2026

ReleasedGPT Image 2 (gpt-image-2) is the first mainstream image model to 'think' before it draws: planning layout, optionally searching the web, and self-checking its output. It tops the LMArena and Artificial Analysis text-to-image boards and replaced DALL·E as the default across ChatGPT and the OpenAI API.

OpenAI's reasoning image model — the most obedient to complex prompts and text.

At a glance

	GPT Image 2
Developer	OpenAI
Released	Apr 21, 2026
Model ID	gpt-image-2
Max resolution	Up to 2K
Aspect ratios	3:1 (ultra-wide) to 1:3 (ultra-tall)
Batch	Up to 8 coherent images per prompt
Text accuracy	~99% char-level (Latin, CJK, Hindi, Bengali)
Pricing (1024²)	~$0.006 low · $0.053 med · $0.211 high

What it is

OpenAI's first image model with native reasoning: it plans a layout, can pull references from web search, and self-checks before returning an image. That reasoning is why it opened a large Elo lead on the blind-vote arenas.

It replaced DALL·E 2 and 3 (both retired May 12, 2026) as the default image model in ChatGPT and the API. Two caveats: GPT Image 2's text-to-image endpoint has no transparent-background parameter and no input-fidelity control — for transparent PNGs or fine edit fidelity, the legacy gpt-image-1.5 is still the better pick.

What it's best at

Prompt obedience and text. It's the model to reach for when the brief is exact — right objects, right positions, nothing extra — or when legible typography across multiple font weights matters, such as magazine layouts and hero brand assets.

What it costs, per channel

Every channel bills GPT Image 2in its own unit. These are their published rates, all reduced to one 1024×1024 image and re-checked daily; the row renders empty until the next sweep covers this model.

Frequently asked questions

GPT Image 2 vs GPT Image 1.5 — which should I use?: GPT Image 2 is the reasoning, quality-first default and wins on prompt adherence and text. GPT Image 1.5 (Dec 16, 2025) remains available as a stable legacy option and is better if you need transparent backgrounds or input-fidelity control, which GPT Image 2 dropped.
Does GPT Image 2 output transparent backgrounds?: Not on its text-to-image endpoint — requests with background: transparent aren't supported. For transparent PNGs, product cut-outs, logos or UI assets, use gpt-image-1.5.
Is 'GPT Image 2' the official name?: Yes. The official API model name is gpt-image-2, released April 21, 2026. OpenAI has not used 'DALL·E 4' — that is not an official name.