An estimate, then a charge.

You choose low, standard, or high. The estimate depends on length and quality. The amount you pay is settled when the run finishes, with a minimum per run.

No API key required. Send either text or word_count, and an optional quality. Prose words are counted the same way a run will bill them: Markdown structure and code are left out.

curl https://humanizeword.com/api/v1/estimate \
  -H "Content-Type: application/json" \
  -d '{"text": "Your draft.", "quality": "standard"}'

{
  "is_estimate": true,
  "prose_word_count": 2,
  "quality": "standard",
  "estimated_usd": { "typical": "0.050000", "high": "0.050000" },
  "hold_usd": "0.050000",
  "basis": "median and p90 of observed cost per 1,000 words"
}

typical is the middle of recent runs at that quality. high is the upper figure the hold is sized against. Both are quoted in USD.

Published prices, with no authentication. Figures are USD per 1,000 prose words. min_charge_usd is the floor for a run.

{
  "model": "usage_based",
  "currency": "USD",
  "min_charge_usd": "0.050000",
  "tiers": [
    {
      "name": "low",
      "charge_per_1k_words_usd": "0.228",
      "high_charge_per_1k_words_usd": "0.366",
      "sample_size": 4
    }
  ],
  "tier_version": 2
}

The live response includes every tier. sample_size is how many runs that figure is based on. A small sample is an order of magnitude, not a fixed rate. Read this endpoint rather than hard-coding the numbers above.

Personas, tones, and the formality range a run can use. Also public. Any option you leave unset on POST /v1/runs is sampled from these lists.

Limits