EvaEsi Reasoning Verify

Overview

Verify does one narrow job: you give it a claim, it goes to the evidence, and it tells you what holds up. Crucially it reports what it could not support as well as what it could. An unsupported claim comes back marked unsupported rather than quietly rephrased into something defensible. It is the model behind Index's citation checking, and it is priced below the deeper Reasoning versions because a narrower job is a cheaper one.

How it behaves

Checks a claim rather than composing an answer.

Speed and depth trade against each other across a family. The faster versions cost more per rung, not less: what is being bought is a shorter wait, and reserving capacity to answer promptly costs more than using it whenever it is free.

Choosing this model

Reach for it when

• Fact-checking claims against sources
• Citation and quotation verification
• Auditing content produced by another model

Reach for something else when

• Use reasoning when you want an answer composed, not a claim checked.
• Use instant when you need the source found rather than assessed.

Others in the Reasoning family

• EvaEsi Reasoning: Deeper synthesis for questions that need working through.
• EvaEsi Reasoning Deep: A longer deliberation budget and a wider evidence sweep.
• EvaEsi Reasoning Fast: Reasoning held to an interactive latency budget.

Example

curl https://evaesi.dev/api/v1/responses \
  -H "Authorization: Bearer esi_dev_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "reasoning.verify",
    "prompt": "Explain what this service does."
  }'

Pricing

Published rates for this model, read live from the platform. API usage is billed at these rates; subscription plans draw from a monthly allowance instead, shown below.

Limits

Plan terms

Subscription plans include a monthly rung allowance on each model, and apply a multiplier when drawing that allowance down. API usage is billed at the rates above and is not drawn from a plan allowance.