.webp)
Krea says its Krea 3 Ultra image editing model has reached the top five of Arena.ai's Multi-Image Edit leaderboard, placing it ahead of models from Google, Meta and ByteDance.
The result puts a smaller independent AI lab alongside much larger technology groups in a benchmark that uses blind head-to-head voting. Arena.ai presents users with two anonymous edits generated from the same prompt and asks them to choose the preferred result. Those votes are then converted into a score similar to an Elo ranking system.
According to Krea, Krea 3 ranks fourth on the leaderboard, with only three OpenAI models above it. It is ahead of Google's Nano Banana Pro and Nano Banana 2.1, ByteDance's Seedream 5.0 Pro and Meta's Muse Image.
The San Francisco-based group has about 35 employees and has raised more than USD $80 million. Krea says that makes the ranking notable because most other leading models in the category come from companies with much larger research teams and market values in the hundreds of billions or trillions of dollars.
Speed focus
Krea is also using the ranking to support its claim that image editing models no longer need to force a trade-off between output quality and response times. The company says Krea 3 can produce a 2K edit in under eight seconds in its fast mode, compared with about 30 seconds for Nano Banana Pro and about 40 seconds for GPT Image 2.5.
That matters because speed has become a practical issue for creative and marketing teams using generative AI tools in day-to-day production. Slower systems can interrupt design workflows when users need to test several image variations, while faster turnaround allows more iterations in less time.
Justine Moore, Partner at a16z and one of Krea's investors, linked the performance claims to changing user expectations around image generation tools.
"For years, image models have made creators choose between the best quality and the speed to actually iterate. Krea 3 removes that tradeoff. When a top-quality edit comes back fast enough to try the next idea right away, the model stops being a step in the workflow and becomes the workflow," said Moore.
Small lab
Krea is positioning itself as an example of how specialist AI developers can still compete in parts of the generative AI market without matching the scale of the largest model builders. Its approach, the company says, has centred on training data and training efficiency rather than the scale of compute or research headcount.
Víctor Perez, Co-Founder and Chief Executive Officer of Krea, argued that anonymous testing offers a cleaner measure of product performance because users judge outputs rather than brand names.
"In a blind test, nobody knows who made the image, so the only thing that matters is whether it's better. Image AI is wide open to disruption. A small team that gets its data right and trains efficiently can go much further than people think, even against companies with thousands of researchers," said Perez.
Commercial use
Krea is pitching the product to companies using image editing in eCommerce, film and advertising. It says Krea 3 supports selected-region editing, bounding boxes for precise placement of elements, and edits based on up to eight reference images. Enterprise customers can also license the model weights and run the system on their own infrastructure, which Krea says is unusual among the highest-ranked editing systems.
According to the company, the software works at native 2K resolution with support up to 4K and is priced at USD $0.06 per image. Users can preserve products, characters and small text while changing other parts of an image, such as materials, outfits, lighting and settings.
Krea says LoRA training and style-reference functions are due to be added, allowing customers to tune the model to a particular product, person or visual style. That would target commercial users seeking consistency across a series of generated or edited images, such as retailers producing product catalogues or marketers adapting campaigns for different formats.
Helena Grau, Head of GTM at Krea, said speed has been the main barrier to wider commercial use of editing models.
"Editing models are already used across eCommerce, film and advertising, but they've been held back by speed. The models good enough for production are too slow for real workflows. Krea 3 brings high-quality edits in under 8 seconds at 2K, and it's the first edit model you can train on a specific product or person. That's what makes it ready for production at scale," said Grau.
Krea says it has 30 million users and that one in three Fortune 500 companies use its platform.