Recraft drops V3, topping image generation benchmarks
Recraft just launched Recraft V3, claiming the #1 spot on Artificial Analysis's text-to-image leaderboard with an 1172 ELO rating, surpassing Midjourney and OpenAI. The new model lets creators place and scale exact text within designs, handles long-form copy, and natively generates vector graphics. It is available immediately for free and paid tiers across web, mobile, and API.

For the longest time, commercial designers using AI image generators had to accept two frustrating realities: gibberish text and zero control over layout.
Design-focused startup Recraft just released Recraft V3 on October 30, and it is taking aim squarely at those exact headaches.
In the Hugging Face Text-to-Image Benchmark run by Artificial Analysis, Recraft V3 climbed to the number one spot after just four days on the leaderboard. It secured an ELO rating of 1172, surpassing established heavyweights like Midjourney and OpenAI.
Long text and layout control
Recraft V3 brings across-the-board upgrades, but its biggest breakthrough is in text rendering. The company claims it is the only model in the world capable of generating images with long passages of text, rather than just isolated words.
Crucially, users are no longer leaving layout to chance. The model allows creators to specify exact text sizes and positions within a design, making it possible to arrange text alongside multiple image elements for complex graphic assets.
Anatomy was another major training target. The model specifically tunes hand counts, limb proportions, and background object positioning relative to the subject. Prompt adherence also got sharper, accurately keeping track of object counts, colors, and relative positions.
Built for professional design workflows
While general image models stop at exporting flattened rasters, Recraft keeps its focus on actual production pipelines.
Its standout feature remains native vector generation, spanning everything from clean icon sets to intricate vector artwork. It also handles custom brand styles natively: users can upload reference images to lock in brand aesthetics without needing to retrain or fine-tune weights.
The update includes a full suite of editing tools, featuring inpainting, outpainting, background removal, upscaling, vectorization, and an AI eraser. Recraft V3 is already live across its desktop Canvas, iOS, Android, and via API for both free and paid accounts.
Why it matters
For creators and marketers, Recraft V3 is worth trying immediately. It bridges the gap between raw AI illustration and usable marketing materials by letting users define text placement and export vector files without jumping back and forth into desktop design suites.
For the AI space, the leaderboard shuffle sends a clear message. The competition is moving away from purely chasing aesthetic vibes toward functional precision. Midjourney proved that AI can make stunning art, but Recraft is showing that real-world utility requires control over copy, layout, and clean vector paths.
The real test will be developer and enterprise adoption. While an ELO score of 1172 is impressive, embedding itself permanently into design agencies and automated production pipelines through its API will dictate whether Recraft can truly challenge mainstream foundation models.
Sources (1)
- recraft.ai(opens in a new tab)Originalrecraft.ai
Topic trackerRecraft V3 launch and leaderboard ascent
- Recraft drops V3, topping image generation benchmarksThis story
- Mystery AI image generator Red Panda tops Midjourney on leaderboard