Model card
LightOnOCR-2-1B.
LightOnopen-source1B paramsVision-Language Model (1B params)Apache 2.03 current SOTA
SOTA on olmOCR-Bench (83.2) with only 1B params. 9x smaller than Chandra-9B, 3.3x faster.
§ 01 · Benchmarks
Every benchmark LightOnOCR-2-1B has a recorded score for.
| # | Benchmark | Area · Task | Metric | Value | Rank | Date | Source |
|---|---|---|---|---|---|---|---|
| 01 | olmOCR-Bench | Computer Vision · Document Parsing | arxiv | 89.6% | #1 | — | source ↗ |
| 02 | olmOCR-Bench | Computer Vision · Document Parsing | old-scans-math | 85.6% | #1 | — | source ↗ |
| 03 | olmOCR-Bench | Computer Vision · Document Parsing | tables | 89.0% | #1 | — | source ↗ |
| 04 | olmOCR-Bench | Computer Vision · Document Parsing | multi-column | 84.8% | #2 | — | source ↗ |
| 05 | olmOCR-Bench | Computer Vision · Document Parsing | pass-rate | 83.2% | #2 | — | source ↗ |
| 06 | olmOCR-Bench | Computer Vision · Document Parsing | long-tiny-text | 91.4% | #2 | — | source ↗ |
| 07 | olmOCR-Bench | Computer Vision · Document Parsing | base | 99.6% | #3 | — | source ↗ |
| 08 | olmOCR-Bench | Computer Vision · Document Parsing | old-scans | 42.2% | #4 | — | source ↗ |
Rank column shows this model’s position vs all other models scored on the same benchmark + metric (competitors after the slash). #1 in red means current SOTA. Sorted by rank, then newest result.
§ 02 · Strengths by area
Where LightOnOCR-2-1B actually performs.
§ 05 · Sources & freshness
Where these numbers come from.
paper
8
results
0 of 8 rows marked verified.