Vibe-Eval benchmark for multimodal chat models
A compact set of challenging vision-language tasks with reference answers in four languages.
- Image → text
- Image classification
- English
- Polish
- Chinese
- +1
Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,788 in the catalog so far, with new ones added every hour.
Answered by Gemini 4 · the chat is kept only in this tab
A compact set of challenging vision-language tasks with reference answers in four languages.
A subset of 113 ObjectNet classes overlapping with ImageNet-1k for testing vision model robustness.
Large collection of 25 million image URLs from Common Crawl with license annotations for image generation.
A large corpus of millions of text-image pairs for training and evaluating product search systems.
Curated 3D scenes with prompts, reasoning, and HTML/JavaScript code for generating 3D graphics.
English dataset of NeurIPS 2026 papers with key claims for reproduction.
40,000 images and 360,000 question-answer pairs for training vision-language models on problem-solving.
1 to 10 million before-and-after video pairs with English text instructions for 10 editing tasks.
A large-scale open corpus in Spanish, Portuguese, and English for fine-tuning large LLMs.
High-precision collision meshes for 1,000 indoor scenes for physically accurate navigation and interaction.
500k pairs of page images and synthetic queries in five European languages for multimodal retrieval.
1,000 selected images of various butterfly species with removed background for image generation.
Image dataset with bounding box annotations for car license plates, split into train, validation, and test sets.
Large array of painting images with text metadata for training generative models and classifiers.
A large dataset of captioned images under a free license for training generative models.
Geospatial dataset and images for evaluating LLM agents.
A dataset with real products, images, and descriptions for category classification.
100 scientific tasks from recent papers to test AI's ability to rank candidates by real experiment outcomes.
Evaluation set for comparing OCR and text/JSON extraction by multimodal LLMs like gpt-4o and gemini-2.0.
An updated RefCOCO validation split in English with corrected masks and cleaned text queries.
A multilingual corpus of image captions translated from English into 200 languages by the NLLB-200 model.
Largest combined dataset of plankton photographs with taxonomic labeling.
An English multimodal benchmark for evaluating whether models can write working 3D scene code.
A set of 2,360 pages of Chinese school exams with question boundary annotations across 9 subjects.