lightonai/LightOnOCR-2-1B
Primitive: /extract · Extract ·
LightOnOCR
Best OCR model . LightOnOCR-2-1B is LightOn's flagship OCR model, refined with RLVR training for maximum accuracy. We recommend this variant for most OCR tasks.
MultimodalMultilingualLong contextEntities
Overview
Hardware: — drives latency, throughput & cost
| Size | 1.0B params |
|---|---|
| Tasks | /extract |
| License | apache-2.0 |
| Languages | en, fr, de, es, it, nl, pt, sv, da, zh, ja |
| Latency | 20.8 s |
| Throughput | 579 tok/s |
| Cost | $0.384 /1M tok |
Cost is approximate — computed from list GPU prices; your actual price depends on the provider you deploy SIE with.
Extraction
| Output kinds | Entities |
|---|---|
| Inputs | image |
| Max sequence length | 16,384 |
Benchmarks
olmOCR-Bench
Document text extraction accuracy across arxiv math, old scans, multi-column layouts, and tables
Corpus: 1,403 Queries: 1,403
default
Quality
accuracy 0.7770
Performance L4 b1 c4
Extract 0.3 mpix/s
Extract p50 20.8s
sglang
Quality
accuracy 0.7782
default_limit-50
Quality
accuracy 0.6743
Compare (0)Compare →