Open
- Rune-1 Small weights on Hugging Face
- Open licence, commercial use allowed
- RuneBench and model cards
- Community support on GitHub
Every account gets 2,500 free credits. Preview models cost 2–3 credits a page. Nothing is billed during the preview.

New accounts get 2,500 free credits. The preview models cost 2–3 credits a page, and nothing is billed during the preview.
Anyone can download, inspect and run our compact models.
Researchers and developers try Rune on their own pages and on RuneBench, in public.
Teams that need higher volume, support or the largest models pay for the API, licences, the app and services.
That money pays for training the next Rune.
During the preview, new accounts get 2,500 free credits and nothing is billed. Try the playground, then call the API from your own code.
# one page in, Markdown out curl https://api.runemic.com/v1/ocr \ -H "Authorization: Bearer $RUNEMIC_KEY" \ -F "file=@page.png" # → { "text": "# Invoice No. 4821 …", "usage": { "credits_used": 2 } }
Credits per page for Rune-1 will be published when it launches. Academic, non-profit and cultural-heritage projects get a discount.
Every new account gets 2,500 free credits. Each page costs a fixed number of credits, depending on the model. Failed pages cost nothing, and nothing is billed during the preview.
| Model | Credits | Pages |
|---|---|---|
Gemma 4 Googlepreview-gemma-4 | 2 | 1,250 |
Llama 4 Scout Metapreview-llama-4-scout | 2 | 1,250 |
Mistral Small 3.1 Mistral AIpreview-mistral-small | 3 | 833 |
GLM-OCR Zhipu AIpreview-glm-ocr | 2 | 1,250 |
LightOnOCR-3 LightOnpreview-lightonocr-3 | 2 | 1,250 |
dots.mocr rednote hi labpreview-dots-mocr | 2 | 1,250 |
KDL Frontier Parser nano KoreaDeeppreview-kdl-parser-nano | 2 | 1,250 |
Credit packs and their prices come later. We will announce them before anything is charged.
Rune-1 Small isn't out yet. When it is, it will be released under an open licence that allows commercial use, published with the weights on Hugging Face.
Rune-1 is multilingual and trained for scripts that most OCR handles poorly: Arabic, Persian and Urdu, Devanagari, Ge'ez, Tibetan, Chinese and Japanese, and historical handwriting, alongside Latin. The exact list, with a production, beta or research status for each script, is published with each release.
Rune-1 Small is compact enough to run on your own hardware. Rune-1 is our largest model, more accurate on complex layouts, tables and rare scripts. It is available through the API and enterprise licences.
No. Documents sent to the API and enterprise deployments are not used for training. For custom models, we only train on material you give us for that purpose.
Rune-1 is in training. Join the early-access list and we'll write to you when the first release is ready. Follow progress on the roadmap.
Yes. Tell us what you have, roughly how many pages, and how fragile it is, and we will plan the project with you.