File Formats · Open Table Formats
Lance
Judgement: Assess
AI/ML-optimised columnar format; trajectory good, adoption concentrated.
Modern columnar file format optimised for AI/ML workloads: 100x faster random access than Parquet, native support for embeddings and multimodal data (images, video, audio), and zero-copy schema evolution without rewriting files.
Pairs with the Lance table format and catalog spec to form a lakehouse stack purpose-built for vector search and feature engineering alongside SQL analytics.
Why it counts as a standard
Lance's on-disk format is published as an open spec with independent Rust and Python implementations. LanceDB, DataFusion, DuckDB integrations, and downstream tools target the format itself rather than a single library — putting it on the de-facto standard track under an open, LF-style community governance model.
Example
A columnar format built for random access: write once, then point-query rows without a full scan.
import lance
import pyarrow as pa
tbl = pa.table({
"id": [1, 2],
"embedding": [[0.1, 0.2], [0.3, 0.4]],
})
lance.write_dataset(tbl, "/tmp/embeddings.lance")
dataset = lance.dataset("/tmp/embeddings.lance")
At a glance
- Category
- File Formats, Open Table Formats
- Governance
- Lance community (independent open source, LF-inspired governance)
- Status
- Emerging; production use concentrated in AI/ML stacks
- First released
- 2022
Links
Related standards
Other standards in File Formats and Open Table Formats.
- AVRO — Apache Avro
- CSV — Comma-Separated Values (RFC 4180)
- Iceberg — Apache Iceberg
- JSON — JavaScript Object Notation
- PARQUET — Apache Parquet
- XML — Extensible Markup Language
- YAML — YAML Ain't Markup Language
- Delta — Delta Lake
See Lance in context
Open the interactive Data Landscape to compare Lance against every other open standard, or grab the raw JSON. Spotted something wrong? Open an issue.