Lenz

Lenz is a fact-checking and hallucination-detection API for teams shipping AI products where a wrong answer is a real liability. Instead of asking one model whether a claim is true, it runs the claim through an independent multi-vendor pipeline: eight models across five stages, grounded in retrieved sources, with an opposing-side debate and three independent reviewers before a verdict is issued. Every result carries its citation trail, so the reasoning can be audited rather than trusted.

The API is built as a depth ladder of four primitives, which lets you match cost and latency to the situation. /extract pulls the verifiable claims out of a block of text. /assess returns a fast multi-model verdict quick enough for synchronous UX. /verify runs the full adversarial pipeline with citations. /ask handles follow-up questions grounded on a completed verification. Typical uses are incident triage, asynchronous runtime verification of model output, and pre-release CI gates on generated content.

It ships an OpenAPI 3.1 spec with Python and TypeScript SDKs and a CLI, and runs inside Claude, Cursor, n8n and Zapier. A browser-based workbench and a bookmarklet cover one-off checks without writing code.

Lenz also publishes its own open evaluation — a study of how much frontier LLMs disagree with each other on real user-submitted fact-checks, with frozen snapshots and full methodology — which is a more useful credibility signal than a marketing benchmark.

Pricing is self-serve with no card required. The free tier gives 1,000 extractions/day, 100 fast checks, 10 deep verifications and 20 follow-ups per month. Plus is $7.99/month, Developer $99/month (5,000 assessments, 500 verifications), and Scale $399/month (20,000 assessments, 2,000 verifications). Enterprise adds SLAs, white-labelling and custom integration support. Annual billing saves two months.