The Moment Everything Changed
The week OpenAI shipped GPT-4o's native image generator in March 2025, Sam Altman tweeted that the company's GPUs were "literally melting." Within seven days, 130 million people had made over 700 million pictures. You could feel it: every tech conference badge, every Substack header, every memed restaurant menu suddenly had the same warm-paper, hand-painted, eerily readable look.
Then, in April 2026, OpenAI replaced it. The new model — gpt-image-2 — landed at #1 on every category of the LM Arena image leaderboard with an Elo of 1512 in text-to-image and 1513 in editing.
130M
People
Used GPT-4o image gen in first week
700M
Images
Created in the first 7 days
1512
Elo Score
gpt-image-2 on LM Arena leaderboard
242
Elo Gap
Lead over next-best model at launch
On LM Arena, the next-best model, Google's Nano-banana-2 (Gemini 3.1 Flash Image), trailed by 242 Elo points at launch — roughly an 80% win rate in head-to-head blind testing. Other public boards show a tighter spread, but the directional verdict is the same: it's not a step. It's a gulf.
The Old Way: Sculpt the Picture Out of Noise
For most of the last five years, "AI image generation" basically meant diffusion. Stable Diffusion, Midjourney, Adobe Firefly, the original DALL·E 2 and DALL·E 3 — they all worked roughly the same way. You start the model on a screen of pure television-static random noise. You hand it a prompt. And then it asks itself, over and over, the same question: given everything I know about real images, what flecks of noise should I subtract right now to make this look slightly more like the thing the prompt described?
Repeat that thirty or fifty times and the static resolves into a cat, a city, a corporate logo. Diffusion is a strange, beautiful, and deeply iterative process, sculpting an image out of fog. It made the modern wave of generative art possible — the canonical paper, "Denoising Diffusion Probabilistic Models" by Ho, Jain & Abbeel, dropped in 2020.
Diffusion has critical weaknesses: garbled text, mangled hands, prompts that ask for "six apples" and produce eight, or three, or somehow a pear. The reason is that diffusion doesn't really plan. It paints. It has no separate, structured representation of discrete objects — it just shapes pixels until the pixels feel right.
This iterative denoising approach was revolutionary for its time — but its fundamental architecture made certain problems nearly impossible to solve.
The New Way: Write the Picture, One Piece at a Time
GPT-4o's image generator in 2025, and now gpt-image-2, do something fundamentally different.
They treat an image the way GPT-4 treats a sentence: as a sequence of tokens to be predicted, one after another. Each "visual token" is a compressed chunk of image content; the transformer stares at the prompt and at all the tokens it has already written, and predicts what the next one should be. Then a small diffusion decoder at the end paints those tokens into actual pixels.
This hybrid is sometimes called the Transfusion architecture in the research literature (Zhou et al., Meta AI, 2024). It is the central conceptual shift — allowing an image model to use the same machinery that makes language models good at logic, counting, and following multi-step instructions.
"The architecture was revamped from scratch." — OpenAI's gpt-image-2 research team
That phrase is not marketing fluff. It explains the leaderboard. The model can now reason about its own output before any pixel exists.
What "Thinking" Buys You
The headline feature of gpt-image-2 is that it is the first image model with native reasoning. You can dial it: low, medium, or high. This sounds like a small thing. It is not.
Plans Composition
At higher reasoning settings, the model plans the full composition before generating a single pixel.
Counts Objects
Counts the objects it has committed to and checks them against the prompt's constraints mid-generation.
Web Search
If it needs current information, it can run a web search mid-generation to stay accurate.
Renders Text
95%+ text accuracy including Cyrillic, Chinese, Japanese, and Korean on curved surfaces and dense layouts.
Older diffusion models start to fall apart somewhere around five to eight distinct objects in a scene; GPT-4o's image stack pushed that ceiling to ten or twenty. The other spec jumps follow from the same core change.
1
gpt-image-1
1024px max, 3 aspect ratios, April 2024 knowledge cutoff, 1 image per call
2
gpt-image-1.5
Interim release: resolution, aspect-ratio, and multilingual-text gains
3
gpt-image-2
2000px max, 7 aspect ratios, Dec 2025 cutoff, 10 consistent images per call, native reasoning
At the API level, a high-quality 1024×1024 image runs roughly $0.21 — meaningfully more than predecessors, which the people running it can charge because the output is roughly twice as good and much harder to redo by hand.
About the Author
David Proctor
David Proctor is VP of AI at Trilogy. He writes about AI infrastructure, agent protocols, and what actually works in production.
AI Infrastructure
Agent Protocols
Production AI