AI Agent StoreLightOn · Document extraction checkpoint
LightOnOCR-3 0.8B
Read page images into text, tables and labeled layout regions. This is the smallest checkpoint in the family.
Announced 2026-10-08 · First reviewed 2026-10-10 · Checked 2026-10-15. The announcement and discovery dates do not establish ordinary API availability. Unverified prices remain unknown.
We still need comparable independent tests. You can compare the confirmed features and limits below, but we cannot yet give this version an Agent Fit or Practical Value rating. Try it on your hardware and tasks before relying on it.
What you can use today
You download the model files and supply the hardware and software to run them. Capacity is the amount of information a request can contain; settings and available memory can limit what fits.
Exact checkpoint ID: lightonai/LightOnOCR-3-0.8B. Use an empty prompt for transcription or the grounding prompt for labeled bounding boxes, image descriptions and chart data. Other instructions are outside the documented training distribution; it is not a general chat or tool assistant.
The card declares Apache 2.0. The 0.8B and 4B variants use Qwen3.5 vision-language architecture, while 1B keeps the previous architecture. Provider OCR tables and normalization choices are not imported as independent general-purpose scores.
The card’s sample output budget is not a published maximum. Image resolution, page complexity and output length change memory use and throughput; the supported document/request cap is unverified.
Hosted prices and measured costs for your documents are unknown. Compare extraction errors, missing chart values, page latency and total operating cost using the same rendered pages.
Provider release and specifications ↗
Check availability and plans ↗
Test it before switching
Give each model the same examples, tools and definition of success. Record correct outcomes, response time, retries and total cost. Include the time and hardware needed to run it.
Compare caching and long-context costs ↗