AI agent benchmark for desktop multitasking with humans
Environment and execution logs for evaluating AI agents operating a computer alongside a human user.
- Reinforcement learning
- Image text → text
- English
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
Environment and execution logs for evaluating AI agents operating a computer alongside a human user.
Approximately 900K question-answer pairs (40% image-based, 60% text-based) for boosting multimodal model reasoning logic.
Over 1 million illustrations extracted from digitized historical books from the British Library (1510–1900).
A collection of images and short videos with questions from Scale AI to evaluate models' ability to understand hidden context.
Curated dataset of 330 thousand high-quality anime arts with tags and prompts for training image generators.
37.6 million image-text pairs from Wikipedia across 108 languages for multimodal retrieval and training.
A large open dataset of 218M images with expanded descriptions enriched with web facts.
Annotated dataset with photos of early school exam pages for evaluating multimodal OCR and VQA.
400,476 videos with previews, view counts, explicit flags, and captions for video-text models.
Over 115k illustrated pages from historical editions of the Britannica encyclopedia (1768–1929).
Dish images with metadata for food recognition, classification, and nutritional analysis.
Detailed synthetic descriptions of 16.9 million images for training and evaluating vision-language models.
A massive corpus of 57.3M short and long English captions for 10M Flickr images from various VLM models.
An SQLite database of Danbooru2023 tags and posts with a latest ID of 7,866,491 for generative models.
A large open multimodal dataset for training Vietnamese visual question-answering models.
8.56M filtered English text-image pairs from LAION-COCO with aesthetic and watermark scores.
Around 300k training and 5k test questions for long-form video understanding.