LightOn Releases LightOnOCR-3: Open-Weight OCR Models Beat Mistral OCR on Benchmarks
wjb_mattingly · x · 2026-10-09
LightOn released LightOnOCR-3, a family of open-weight (Apache 2.0) document intelligence models in 0.8B, 1B, and 4B sizes (the 0.8B and 4B use the Qwen3.5 VLM architecture):
- One model handles both transcription and layout: an empty prompt transcribes the page; a "grounding" prompt adds labeled bounding boxes, image descriptions, and chart data extracted as tables
- Leads OlmOCR-Bench ahead of ChandraOCR-2, Mistral OCR 4.1, and dots.ocr; first among open-weight models on ParseBench
- Significant speed and transcription quality gains over LightOnOCR-2, with up to 2x document processing throughput
More from Models
- Perplexity Decider tops DecisionBench with 93.9% accuracy, 534ms latency, and lowest cost — AravSrinivas · 2026-10-09
- Claude 3 Opus confirmed working within Max plan monthly API credits — repligate · 2026-10-09
- Leak: Grok Voice Mode coming to X — talk to Grok out loud in the app — nima_owji · 2026-10-09
- Subscription Claude models deliver far fewer thinking tokens, measured five ways — _AustinCalvert_ · 2026-10-09
- GPT-6 in ChatGPT is the best AI writer yet — barely needs editing, says user — VraserX · 2026-10-09
- Reddit User Finds GPT-6 Appears to Lack Direct Access to Saved Memories — ItsAGarbageAccount · 2026-10-09