Observed arrival · 2026-10-19
LDF: a compact table format for language models
LDF converts JSON records into typed, line-oriented tables intended to use fewer tokens when language models read structured data.
- For
- Developers feeding structured records to language models
- Worth noticing
- The benchmark uses o200k_base on the first 1,000 records and publishes misses alongside token ratios.
Field notes
LDF organizes a document as tables with typed headers and rows; stable nested objects can become dotted columns, while arrays of objects use child tables joined through the parent’s first column. Its examples distinguish a missing key from a present null and describe a sparse-key rule for values present in fewer than half of rows. The published benchmark uses o200k_base on the first 1,000 records in its datasets and includes results as well as reported misses.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue