All models
    Frontier · Decision models

    The PurpleMIST family

    Our frontier models. Give them a message, ticket or record and any number of typed questions, and they return calibrated probabilities for every answer in one pass, in English and African languages.

    Meet the models Compare
    / The lineup

    2 models, one interface

    DECISIONSMOST ACCURATE

    PurpleMIST-Flash-1.0

    Frontier-level decisions from a model 200× smaller.

    0.720
    LocalLLaMA/typed-decisions, English zero-shot
    0.668
    African Typed Decisions, 8 languages
    5×
    Closer probabilities than DeepSeek-V4-Pro
    7.9B
    Parameters
    Best for
    • Routing and triaging support tickets, chats and emails in English and African languages.
    • Compliance, moderation and KYC checks where you need a probability, not just a label.
    • Replacing chains of prompted LLM calls with one fast pass that answers every question together.
    Learn more 🤗 Weights
    DECISIONS

    PurpleMIST-Mini-1.0

    PurpleMIST decisions in 1.9B parameters.

    0.601
    LocalLLaMA/typed-decisions, English zero-shot
    0.058
    Calibration error (ECE), same benchmark
    1.9B
    Parameters
    7.5 GB
    Memory in fp32
    Best for
    • High-volume English classification and routing where cost per request matters.
    • Prototyping a decision pipeline before scaling up to PurpleMIST-Flash-1.0.
    • Smaller GPUs and shared inference servers.
    Learn more 🤗 Weights
    / Side by side

    Compare the PurpleMIST models

    PurpleMIST-Flash-1.0 →PurpleMIST-Mini-1.0 →
    Parameters7.9B1.9B
    LocalLLaMA/typed-decisions (English, zero-shot)0.7200.601
    African Typed Decisions (8 languages)0.6680.416
    HardwareOne 24 GB GPUAbout 7.5 GB in fp32
    Best forAfrican languages and top accuracyEnglish at lower cost
    / Choosing

    Which one should you use?

    • Calibrated decisions on messages and tickets, African languages includedPurpleMIST-Flash-1.0 →

      Frontier: matches DeepSeek-V4-Pro accuracy at 7.9B: 0.720 on LocalLLaMA/typed-decisions (English, zero-shot) and 0.668 on African Typed Decisions (8 languages).

    • English decisions on smaller hardwarePurpleMIST-Mini-1.0 →

      Same interface at 1.9B: 0.601 on LocalLLaMA/typed-decisions (English, zero-shot).

    Other model families

    MIST-Extract

    Document extraction · 2 models

    LID-608

    Language identification · 2 models

    DiacNet

    Diacritic restoration · 3 models

    MIST QG

    Question generation · 1 model