Lenz Review (2026): Pricing, Features & Honest Verdict

Lenz is a fact-checking API that runs a claim through an eight-model adversarial pipeline and returns a sourced verdict with citations. We found the engineering honest and the free tier real, but the price per check rises 24 percent at the tier most developers land on. Best for: AI product teams shipping factual content. Price: From $7.99/mo (free plan: yes). Rating: 7/10

What is Lenz?

Lenz launched on Product Hunt on 27 August 2026 at #3 for the day with 173 upvotes. It is built by Lenz IO, and the product is narrow on purpose: it takes one factual claim, researches it from scratch, argues both sides of it with two separate models, has three more models audit the argument, and returns a verdict you can hand to an auditor. Not a confidence score. A verdict with the sources attached.

The framing that makes this worth reading is on their own about page. Five frontier models, each ruling on the same 1,000 real user-submitted claims, disagreed more often than they agreed. On 63 percent of those claims at least one model broke from the panel majority or no majority formed at all, and on 23 percent there was a substantial split, meaning two models landed two or more verdict options apart. That is the whole argument for buying this instead of asking your existing model twice. A single confident answer hides the disagreement. Lenz surfaces it and then settles it on retrieved evidence.

What Are Lenz’s Key Features?

Four API primitives that form a depth ladder

This is the part we liked most, because it respects that most checks do not need the expensive path. POST /extract pulls atomic verifiable claims out of any block of text, dropping opinions and meta-commentary, resolving pronouns, splitting compound sentences. It is free, 1,000 calls a day. POST /assess runs a three-model panel and comes back in roughly 5 to 10 seconds, which fits inside a chat completion timeout. POST /verify runs the full eight-model pipeline with citations and takes about 90 seconds. POST /ask answers follow-up questions grounded on a finished verification, using the same evidence rather than re-researching. The documented pattern is to run /assess in the user-facing path and escalate only the low-confidence claims to /verify in the background, which is the right advice and the cheap one.

A five-stage pipeline that publishes its own working

Every /verify call passes through framing, research, debate, panel review and conclusion. Framing strips emotional language and turns the input into a neutral testable hypothesis with targeted search queries. Research hits multiple independent search engines in parallel, reads the actual pages, extracts quotes from the source text and scores each piece of evidence for authority, relevance and recency. Debate uses two separate models to argue opposing sides across two rounds, each writing a rebuttal to the other. Panel review puts three more models on three different axes: source reliability, whether the evidence logically supports the claim, and whether the claim’s numbers and wording match what the evidence actually shows. Conclusion synthesises all of it into one of five verdicts, True through False, plus a Lenz Score from 1 to 10. The full trace ships with the result.

A free tier that is actually usable in CI

The free plan is not a demo. It gives you 1,000 extractions a day, 100 fast checks a month, 10 deep verifications and 20 follow-ups, with no card required. That is enough to wire /extract into your deploy pipeline and answer the question “what factual claims did our AI make yesterday” for zero spend before you commit a euro. Very few API companies hand you the entry primitive at production volume. This one does.

Integration surface that does not force a rewrite

There is an OpenAPI 3.1 spec at a public URL, Python and TypeScript SDKs, a CLI, and connectors for Claude, Cursor, n8n and Zapier. There is also a browser bookmarklet and a public library of completed verifications you can read without an account, which doubles as the sales pitch and the documentation. The company publishes its own evaluation study with frozen snapshots and full methodology, which is a level of self-scrutiny most tools in this category skip.

How Much Does Lenz Cost?

Lenz starts at $7.99 a month for Plus, with Developer at $99, Scale at $399 and a free tier at $0. Annual billing is $79, $999 and $3,990 respectively. Every tier carries the same clean quota ratio: one deep verification for every ten fast checks and every two follow-ups. Free gets 10 verifications a month, Plus 50, Developer 500, Scale 2,000.

Now the number that should decide your tier, because Lenz does not advertise it. Divide each price by its deep verification allowance and the unit cost is 16.0 cents on Plus, 19.8 cents on Developer and 20.0 cents on Scale. It goes up, not down. The reason is the jump from Plus to Developer: 12.39 times the price for exactly 10 times the volume, a 24 percent increase in what each check costs you. Developer to Scale is honest at 4.03 times the price for 4 times the volume. So the expensive move is the first commitment, and it is the one every serious user has to make, because Plus caps out at 50 checks a month.

The annual discount is small but mostly accurate. The “2 months free” badge would mean paying ten months for twelve. On Scale it is exact, $3,990 against ten times $399. On Plus it beats the promise by 90 cents. On Developer it misses by $9, since ten times $99 is $990 and the annual price is $999. The stated dollar savings on the page reconcile to the cent against real monthly prices, which is more than most pricing pages manage.

One quota never grows. /extract is 1,000 a day on the free plan and 1,000 a day on the $399 Scale plan. In practice that is not the binding constraint, since Scale only includes 20,000 fast checks a month, but it is a strange thing to leave flat across a 50x price range. There is also no published overage rate anywhere. Cross your allowance and your only option is the next tier.

Who is Lenz Best For?

Use Lenz if you ship AI-generated text that a customer, an auditor or a regulator might later question, and you need a third-party record showing the claims were checked before publication. Insurance, legal research, medical content, financial summaries, compliance reporting. The audit trail is the product. The verdict is almost a byproduct.

Use Lenz if you already have a golden test set and want hallucination regressions caught on every deploy. /extract is free at 1,000 calls a day, so that use case costs nothing to start and produces signal on day one.

Skip Lenz if you need a verdict inside a synchronous user-facing request and cannot accept a 5 to 10 second wait. /assess is the fast path and it is still slower than most UI budgets allow. The 90-second /verify is a background job, and the docs say so plainly.

Skip Lenz if your volume sits awkwardly between tiers. At 60 deep checks a month you are paying $99 for a job that costs $9.59 worth of Plus capacity, because there is no overage and no metered option. Anyone doing between 50 and 300 checks a month is overpaying, and that band covers a lot of small products.

Best Lenz Alternatives

Exa is the retrieval layer you would build on if you wanted to assemble this yourself. It starts free with $20 in signup credits plus $10 a month, then Search costs $7 per 1,000 requests, page contents $1 per 1,000 pages and Deep Search $12 to $15 per 1,000 requests. At roughly 1.2 to 1.5 cents a deep search call it is around a fifteenth of a Lenz verification, but you are buying documents, not a verdict. The debate, the panel and the audit trail are yours to build.

Tavily is the other web-access API for agents, with a free Researcher tier at 1,000 API credits a month and pay-as-you-go at $0.008 a credit. Same trade as Exa: cheaper, more flexible, and it does not tell you whether the claim is true. Worth pairing with Lenz rather than choosing between them, since Tavily’s per-credit metering is exactly the pricing model Lenz lacks.

Originality.ai comes at the same problem from the content side rather than the API side. Pro is $14.95 a month, or $12.95 billed yearly for a $155.40 annual total, with 2,000 credits a month and extra seats at $9.95. Enterprise is $179 a month, or $136.58 billed yearly. It is aimed at editorial teams checking human and AI writing, not at a runtime gate inside a product, and it does not return a citation trail you could hand to a regulator.

Final Verdict: Is Lenz Worth It?

Lenz is a well-built product with an unusually honest company behind it. Publishing a study that shows frontier models disagreeing on 63 percent of real claims is a strange marketing decision until you realise it is the argument for the whole category, and Lenz is the only company we have seen make it with data instead of adjectives. The five-stage pipeline is specified in public, every verification ships its own trace, and the free tier is generous enough to prove the thing works before you pay. The 1:10:2 quota ratio holding across all four tiers is a small sign of a team that thought about the shape of the offer rather than picking round numbers.

The pricing is where we would push back. A per-check cost that climbs 24 percent exactly when a developer commits is the wrong shape, and the absence of any overage rate means the ladder has no middle rungs. If your volume is 500 or 2,000 checks a month, Developer and Scale are fairly priced and the audit trail justifies them. If you are at 60 a month, you will pay $99 for $10 of use and feel it. Buy Lenz for the trace, not for the unit economics, and run /extract free in CI for as long as you can before you commit to a tier.