Editorial & Data Methodology

Last reviewed: August 5, 2026. This page explains how AI Agent Hub chooses sources, handles fast-changing data, calculates estimates, and corrects errors.

1. Source hierarchy

  1. Model facts and API prices: provider documentation, model cards, or official API pricing pages.
  2. Open projects and protocols: original specifications, repositories, or maintainer documentation.
  3. Benchmarks: original benchmark reports or provider-published result tables, with attribution.
  4. Secondary sources: used only for context and not as the sole support for a material price or availability claim.

2. Data snapshot rules

Each catalog snapshot records the date checked, token units, input and output rates, context limit, source URL, and special conditions. Temporary prices are labeled with an end date when the provider publishes one. If a provider announces a future pricing scheme without an effective date, we do not apply it early.

3. Calculator rules

The cost calculator is an estimate, not a bill forecast. It multiplies user-entered token volumes by the displayed per-million-token rates. Cache savings apply only to the portion the reader enters as cacheable. The Gemini 3.6 Flash entry currently assumes no cache discount. Taxes, currency conversion, reseller markup, minimum commitments, batch discounts, and account-specific tiers are excluded unless stated.

4. Benchmark rules

The current benchmark matrix contains a deliberately small comparable set. Its SWE-Bench Pro and Terminal-Bench 2.1 values are vendor-published results shown in the Google DeepMind Gemini 3.6 Flash performance table. AI Agent Hub did not run those evaluations. Vendor-reported results may differ in harness, inference settings, or available tools and should not be read as an independent endorsement.

5. Recommendations

We avoid universal “best model” rankings. A useful choice depends on task quality, latency, privacy controls, tool support, region, and workload cost. Where the site offers analysis, it must separate opinion from sourced facts and state important assumptions.

6. AI assistance

AI tools may assist with research organization, drafts, code, and quality checks. AI output is not treated as evidence. The site operator selects sources, reviews claims, decides what is published, and remains accountable for errors.

7. Corrections

Report a suspected error through the contact page with the page URL, exact statement, and supporting primary source. We review high-impact pricing, availability, and security issues first. When a shared data snapshot changes, its verification date is updated.

Current primary data sources