Reve releases Reve 2.1 image model
Independent lab Reve has shipped Reve 2.1 a month after version 2.0, saying the model is back in second place overall on Arena even though it used a fraction of rivals' training compute.
Reve announced Reve 2.1, an image generation and editing model released about a month after Reve 2.0. The company says the update brings a "rapid leap in visual intelligence and world knowledge."
Reve makes three ranking claims. It says the model is back in second place overall, describes it as "the top 4k model in the world," and calls Reve the highest-ranked independent foundation image model lab on both Arena and Design Arena. It also says the model was trained with less than a tenth of the compute used by the labs ranked directly above and below it. The post does not give leaderboard scores, name those labs or report compute figures. It also cites no independent benchmark results.
According to the company, Reve 2.1 makes these improvements:
- Layout planning: the model works out structure, hierarchy and spatial relationships before rendering, which Reve says helps with dense, complicated scenes.
- Precision editing: each element can be addressed, so a single region can be changed and re-rendered on its own.
- Native 4K output: images render at 16 megapixels, and Reve says detail holds up through repeated edits.
- Multilingual text rendering: dense text inside images now extends to foreign scripts.
Reve says the release builds on a bet it made two years ago that images should be "built like code," as hierarchical, structured layouts. Its separate "The Layout Bet" post describes a "large layout model" in more detail.
Reve 2.1 is available now at reve.com. The announcement does not cover pricing, API access or model size.