PatentBench

Patent prosecution AI benchmark from ABIGAIL covering USPTO Office Action parsing, docketing, response strategy, drafting, prior-art analysis, and hallucination checks.

5rows
overall_accuracyprimary metric
2026-05-26sampled

Metadata

ID
patentbench
Category
Legal
Release
2026-03-20

Metrics

Overall Accuracy, Action Classification, Timeline Analysis, Fee Computation, Deadline Calculation

Latest Results

Layer 1 deterministic docketing rows only; Layer 2 scores were pending on the source page.

Rank Subject Overall Accuracy Model Match Provenance Sampled
1 ABIGAIL v3 100 — Imported 2026-05-26
2 Variant B 95.90 — Imported 2026-05-26
3 Claude Sonnet 4 99.10 Claude Sonnet 4
anthropic-claude-sonnet-4
 		Imported 2026-05-26
3 Gemini 2.5 Flash 99.10 Gemini 2.5 Flash
google-gemini-2.5-flash
 		Imported 2026-05-26
5 Gemini 2.5 Pro 88.70 Gemini 2.5 Pro
google-gemini-2.5-pro
 		Imported 2026-05-26