U.S. Patent Grants and Applications from 1976 to 2025

Approximately 8 million U.S. patent grants and applications from 1976 to 2025, cleaned and formatted for language model pre-training. The dataset was created by AllenAI and last updated on December 18, 2025.

Use Cases

• Pre-training language models based on the large corpus of technical and legal text.
• Analyzing patent filing trends over time based on the filing_date field.
• Classifying patent documents by type based on the patent_type field.
• Studying the evolution of technical language based on the concatenated title, abstract, and specification text.

Strengths

• Approximately 8 million records, providing substantial scale for model training.
• Covers a 50-year time range from 1976 to 2025, enabling longitudinal analysis.
• Data is described as cleaned, filtered, and formatted for a specific use case.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is approximate, and exact file size and formats are unknown.

Provenance

Source

AllenAI via Hugging Face.

Collection Method

Collection of U.S. patent grants and applications, cleaned and filtered.

Time Range

1976-2025

Freshness

Last updated 2025-12 18 23:35:39.

Geography

United States