GPTKB v2
A large general-domain knowledge base entirely from a large language model.
Browse all Dataset profile SPARQL query Download
38.5 million triples 1.3 million entities
96% entity precision 100% homonymy 91% synonymy resolved
About
GPTKB v2 is a general-domain knowledge base built from GPT-5 — 38.5 million triples across 1.3 million disambiguated entities. The browser exposes the disambiguation chain (surface form → canonical entity → provenance batch) on every triple. See Hu et al. (2026) for how it is built.
Try an entity
Example SPARQL queries
Main papers
The Publications page lists the earlier GPTKB work this builds on.
GPTKB 2.0: Direct Construction of Disambiguated Knowledge Bases from Large Language Models
arXiv, 2026
GPTKB 2.0: Browsing, Querying, and Auditing a Disambiguated LLM-Derived Knowledge Base
arXiv, 2026