Mystery AI image generator Red Panda tops Midjourney on leaderboard

A mystery contender just showed the AI image heavyweights how it's done.

On the blind benchmark platform Artificial Analysis, an unannounced model dubbed 'Red Panda' suddenly surged past established giants like Midjourney v6, DALL-E 3, Stable Diffusion 3 Large, and FLUX.1 [pro]. Days later, design startup Recraft stepped forward to reveal that Red Panda was actually its newly released Recraft V3 model.

Dominating the blind arena

Artificial Analysis evaluates text-to-image models through head-to-head blind tests. Users are presented with two unlabeled images generated from the same prompt and vote for the best rendition based on image quality, prompt fidelity, and artifact absence.

Red Panda racked up a dominant 72% win rate and an Elo rating of 1172.

That performance placed it 40 points ahead of the runner-up, Black Forest Labs' FLUX 1.1 [pro], the engine powering Grok's image generation on X. Testing the model over 60 rounds, ZDNET confirmed Red Panda's strength, noting it tied with Midjourney and FLUX.1 as top performers.

Dropping a mystery model into an arena before launch has become a standard playbook for labs looking to build authentic buzz and benchmark raw performance before an official rollout.

Built for designers, not just art prompts

With the reveal, Recraft detailed what makes V3 distinct: it is engineered directly for graphic design and vector workflows.

Text rendering—long a weak spot for diffusion models—is front and center. Recraft claims V3 delivers state-of-the-art text generation capable of producing accurate text of any size and length directly inside the artwork.

Beyond typography, Recraft V3 improves human anatomy realism, positional spacing between scene elements, and overall prompt adherence. The broader platform wraps this engine with functional tooling, including custom brand color palettes, positioning controls, and collaborative workspaces.

Why it matters

For graphic designers, marketers, and product builders, Recraft V3 targets the friction of turning raw AI outputs into usable assets. Platforms like Midjourney produce stunning visuals, but users typically have to pull those assets into other software to fix butchered typography or enforce brand colors. Recraft integrating vector outputs and reliable text rendering makes it immediately worth testing for everyday mockups and social collateral.

Strategically, Recraft made a smart pivot by competing on precision rather than photorealism. While the rest of the industry races toward general visual fidelity, anchoring a model to real-world design constraints gives Recraft a practical defensive moat. That said, arena test scores reflect prompt enthusiasm more than daily grind reliability. Whether Recraft V3 can maintain its edge under dense, multi-layer commercial production remains to be seen.