All models
    Specialist · Document extraction

    The MIST-Extract family

    Give it a JSON template and a document as text, a photo or both, and it fills in the template with a calibrated confidence for every field. Built for receipts, mobile-money messages, bills and letters in English and African languages.

    Meet the models Compare
    / The lineup

    2 models, one interface

    EXTRACTIONMOST ACCURATE

    MIST-Extract-1.0

    Documents in, JSON out, with a confidence on every field.

    91.8%
    English documents fully right, held-out
    0.728
    Named entities, 21 African languages
    0.008
    Calibration error (ECE) on field confidence
    4B
    Parameters
    Best for
    • Mobile-money and bank messages: transfer confirmations, wallet SMS, alerts and statements.
    • Receipts and bills in naira, cedis, shillings, rand and other African currencies, from text, photos or screenshots.
    • Letters, forms, IDs and tickets, with names of people, places and organisations in 20+ African languages.
    Learn more 🤗 Weights
    EXTRACTION

    MIST-Extract-Flash-1.0

    MIST-Extract at 2B, for documents at volume.

    90.5%
    English documents fully right, held-out
    0.701
    Named entities, 21 African languages
    9.7/s
    Documents per second, one RTX A6000
    2B
    Parameters
    Best for
    • High-volume receipt, bill and payment-message ingestion where cost per document matters.
    • Receipt photos in African currencies and layouts.
    • Prototyping an extraction pipeline before moving up to MIST-Extract-1.0.
    Learn more 🤗 Weights
    / Side by side

    Compare the MIST-Extract models

    MIST-Extract-1.0 →MIST-Extract-Flash-1.0 →
    Parameters4B2B
    English documents fully right (held-out)91.8%90.5%
    Named entities, 21 African languages0.7280.701
    Field accuracy, all held-out documents0.8980.879
    Speed, one RTX A60006.0 docs/s9.7 docs/s
    Confidence calibration error (ECE)0.0080.022
    Best forTop accuracy, general and African documentsVolume and receipts
    / Choosing

    Which one should you use?

    • JSON from receipts, payment messages and letters, African languages includedMIST-Extract-1.0 →

      91.8% of held-out English documents fully right (NuExtract3: 67.9%), 0.728 on African-language named entities, calibrated confidence on every field.

    • Document extraction at volumeMIST-Extract-Flash-1.0 →

      Same templates and confidence at 2B: 9.7 documents per second on one GPU, 90.5% of English documents fully right.

    Other model families

    PurpleMIST

    Decision models · 2 models

    LID-608

    Language identification · 2 models

    DiacNet

    Diacritic restoration · 3 models

    MIST QG

    Question generation · 1 model