Your data is hierarchical.Your model should be too.

hyper3-clip-v1 is our hyperbolic image–text model for visual search. Search from broad categories to specific descriptions, matching objects, attributes, and relationships.

Request EvalHugging Face LogoGet hyper3-clip-v1

Where exact matches get buried.

Explore saved comparisons of hyper3-clip-v1 and OpenAI CLIP on product, fashion, and object search.

More demos

“a product photo of grey velvet sofa”

Exact query

a product photo of grey velvet sofa: Frederick Mid-Century Modern Tufted Velvet Sofa Couch 77.5 W Grey, with metal, brass finish, wood, velvet upholstery, tufted

Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5 W, Grey

Exact product

Frederick Velvet Sofa

hyper3-clip-v1

Exact product #1
  1. Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5"W, Grey#1

    Frederick Velvet Sofa

    Exact product
  2. Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Charcoal & Brass#2

    Uptown Velvet Daybed

  3. Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5"W, Forest Green#3

    Frederick Velvet Sofa

  4. Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5"W, Navy Blue#4

    Frederick Velvet Sofa

  5. Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Dove Grey & Brass#5

    Uptown Velvet Daybed

OpenAI CLIP

Exact product #5
  1. Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Hunter Green & Brass#1

    Uptown Velvet Daybed

  2. Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Charcoal & Brass#2

    Uptown Velvet Daybed

  3. Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Shadow & Silver#3

    Uptown Velvet Daybed

  4. Rivet Alonzo Contemporary Leather Sofa Couch, 80"W, Grey#4

    Alonzo Leather Sofa

  5. Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5"W, Grey#5

    Frederick Velvet Sofa

    Exact product

See how your own catalog compares.

Get a 48-hour evaluation

Why Geometry Matters

Standard models cram your data into a flat (Euclidean) space. This creates a fundamental mathematical mismatch:
  • Hierarchical data grows exponentially.
  • Euclidean space grows only polynomially.
  • The Mismatch: Flat models force exponential hierarchies into polynomial dimensions, crowding siblings together and causing retrieval failures like category bleed.
Hyperbolic space naturally expands exponentially, giving every variant room to breathe.

Euclidean Space

Overcrowding

Hyperbolic Space

Separation

Results

From matching the details in a description to finding products and objects. Explore the compositional benchmark and our visual-retrieval evaluations below.

Compositional matching

SugarCrepe · macro accuracy ↑

SugarCrepe macro accuracy; higher is better
ModelAccuracy ↑
hyper3-clip-v179.54%
Jina CLIP v178.20%
Jina CLIP v275.02%
OpenAI CLIP ViT-B/1673.06%

Visual retrieval

Image and product retrieval against OpenAI CLIP.

Swipe to compare →

Industry / DatasetBenchmarkhyper3-clip-v1OpenAI CLIPReadout
Ecommerce Catalog Retrieval – Amazon Berkeley Objects
Retail catalogs500 product images, 20 product typesProduct-type mAP0.5820.552+3.05 pts
Retail catalogs50 parsed catalog departmentsDepartment mAP0.2640.212+5.20 pts
Retail catalogsParent category retrieves diverse childrenChild coverage@500.7800.655+12.50 pts
Fashion Matching & Search – DeepFashion In-Shop
Apparel retail710 photo queries, 741 catalog photosmAP0.6350.352+28.3 pts
Apparel retailSame product across views; query photo excludedRecall@10.7590.456+30.3 pts
Apparel retail180 text queries, separate 1,120-photo poolHit@100.5720.550+2.2 pts
General Visual Hierarchy – COCO Objects
Object search5,000 COCO val images, 80 categoriesCategory mAP0.5540.532+2.22 pts
Object search12 object supercategoriesSupercategory mAP0.5360.516+2.08 pts
Object searchCoverage of child types under broad labelsChild coverage@1000.8870.951CLIP +6.40 pts

hyper3-clip-v1

hyper3-clip-v1 brings images and text into a shared hyperbolic space for broad-to-specific retrieval and compositional matching.

Hugging Face LogoGet the model and quick start

HyperView

HyperView is an agent-native workbench for inspecting embedding spaces, curating datasets, and understanding why retrieval results fail.

GitHubView on GitHub

Get a 48-hour retrieval eval.

Send a small sample of your images and queries. We run hyper3-clip against your current baseline and return a short report within 48 hours: metrics, ranked examples, and whether a pilot is worth it. No discovery call required.

01Copy the prompt into your assistant
02Answer three questions, send the email
03Get an eval report back within 48 hours

Our Research

Notes, papers, and technical writeups behind the model.