API Models

    The models available on the Olaverse API (api.olaverselab.com). Pricing is pay as you go in US dollars, with no subscription, and every new account starts with $1 of free credit. All models are also published as open weights on Hugging Face.

    Full reference: docs.olaverselab.com · Pricing: docs.olaverselab.com/reference/billing

    1. Language Identification

    lid-neural-608 · lid-lite-608

    POST /v1/language-id · POST /v1/language-id/batch

    Identifies the language and script of a text from 608 languages. Built African-first, it holds up on short, messy user input such as chat messages, where general detectors often confuse Yorùbá, Igbo or Hausa with other languages.

    • lid-neural-608 is the most accurate; lid-lite-608 (fastText) is the fastest.
    • "traffic" mode suits chat, user input and routing; "coverage" mode treats every language equally, for building corpora.
    • Up to 25 runner-up languages returned with scores.
    Input
    Text, up to 1,000 characters per request. Batch supported.
    Output
    JSON: language code, script, language name, confidence and top-k alternatives.
    Category
    Translation → Language Detection
    Pricing
    lid-neural-608: $0.30 per 1,000 requests
    lid-lite-608: $0.05 per 1,000 requests
    Docs
    https://docs.olaverselab.com/api/language-id/

    Example

    → {"text": "Mo fẹ́ lọ sí ọjà"}
← {"language": "yor", "script": "Latn", "language_name": "Yoruba", "confidence": 0.97, "top": [...]}

    2. Diacritic Restoration

    diacnet-2.0 · diacnet-mini-2.0 · diactag-2.0

    POST /v1/diacritize

    Restores missing diacritics, tone marks and Arabic vowel marks in text typed without them. This matters for search, text-to-speech, translation quality and readable user content.

    • Languages: Yorùbá, Igbo, Hausa, Arabic, Vietnamese, Polish, Turkish, Portuguese, Spanish, French and Italian, or automatic detection.
    • diactag-2.0 never changes a letter and can leave low-confidence characters exactly as typed.
    • Batch supported, with one language per text.
    Input
    Text.
    Output
    JSON: the restored text.
    Category
    Text → Spell Check / text correction (closest match)
    Pricing
    diacnet-2.0: $1.00 per 1M characters
    diacnet-mini-2.0: $0.60 per 1M characters
    diactag-2.0: $0.20 per 1M characters
    Docs
    https://docs.olaverselab.com/api/diacritics/

    Example

    → {"text": "se eranko naa si gbo o?", "language": "yor"}
← {"text": "ṣé ẹranko náà sì gbọ́ ọ?", "model": "diacnet-2.0", "language": "yor"}

    3. Document Extraction

    mist-extract-1.0

    POST /v1/extract · POST /v1/extract/jobs

    Extracts fields from a document into the JSON template you supply, with a confidence score for each field. Strongest on receipts, invoices, forms and short business documents.

    • Works on text, images (JPEG/PNG) and PDFs.
    • Long documents run as asynchronous jobs, with a cost estimate up front.
    • Unparseable extractions are not charged.
    Input
    Document (text, image or PDF) plus a JSON template.
    Output
    JSON matching the template, with per-field confidence.
    Category
    OCR → Custom Document Parsing / Data Extraction
    Pricing
    $0.80 per 1M input tokens, $4.00 per 1M output tokens
    A one-page text receipt costs about $0.0015
    Docs
    https://docs.olaverselab.com/api/extraction/

    Also available

    Question generation (mist-qg-1.5b) and a structured decision model (purplemist-mini-2.0). See the model reference.

    Data handling: inputs are processed in memory and never used for training; extraction job results are kept for 7 days. See our Privacy Policy.