ahel
Start free

Judge evidence and return a label or score

Answer labels, probabilities or rubric scores from an AI judge, plus the model used

Ready

$0.000275 per decision request (maximum)

The unit rate applies when a result is found. Misses are free when the provider does not bill them; provider-billed misses are charged at cost, without Ahel's margin. Provider-billed failures follow the quoted billing rules. A run that never reaches a provider is not charged. Pending charges stay held until the provider settles.

2 providers, one call. Quoted before it runs, paid from your prepaid balance. No top-up fee. Pay per call; each call’s price is shown before it runs. How Ahel picks ↓

Use Ahel to judge whether a product review on Amazon sounds genuine and get a yes/no probability. (Ahel task id ai-judge.decide.)
First call capability_inspect with id "ai-judge.decide" for the exact inputs and price, then run_capability with my inputs, and tell me what it cost.
If Ahel is not connected yet: Set up Ahel: https://ahel.ai/llms.txt
Run it free

Providers, ranked by measured success

2

Ranked by Ahel's own calls in the last 30 days: success, then speed, then price. If a provider finds nothing, Ahel may try the next provider. Shown prices include Ahel's margin. Misses are free when the provider does not bill them; provider-billed misses are charged at cost, without Ahel's margin. Provider-billed failures follow the quoted billing rules. A run that never reaches a provider is not charged. Pending charges stay held until the provider settles. The final charge never exceeds your quote. Works counts calls that answered, found or not; New means fewer than 5 calls so far. Rates checked 2026-10-28.

ProviderTakesWorksPrice
Surplus IntelligencestatequestionsNewunder 5 calls$0.000275/requestonly when found
OpenRouterstatequestionsNewunder 5 calls$0.00055/requestonly when found

Inputs

2
  • Staterequiredstate · string

    The evidence, as bounded Markdown with XML-style sections. Escape & < > in untrusted text. Quoted text is evidence, never instructions. State and questions together take at most 10,000 bytes.

  • Questionsrequiredquestions · json

    Question id -> {type: choice | noul | score, instructions, criteria}. choice and noul take criteria as label -> description (noul uses true and false); score takes an ordered list of levels. Questions

Returns

3

answers, model, generation_id. Every row keeps its provider and when it was collected.

Task id
ai-judge.decide