Fresh tasks. Cross-company judges. Published evidence.

Trust ratings AI companies can't game.

Benchmarks are gameable, and a leaderboard score doesn't tell you how a model behaves on your actual work. Algodai tests AI models and products on fresh, rotating real-world tasks — and every transcript is judged by a rival company's model, never the maker's own.

An AI model isn't a frozen artifact like a smart contract — it gets updated, fine-tuned, and outpaced by the next release. So a one-time certificate is the wrong shape of trust. Algodai re-tests over time and shows current competitive standing, not just a score frozen on day one.

Never self-graded · Task pool rotates every cycle · Every claim links to an inspectable artifact

Security

Prompt injection, jailbreaks, tool abuse, data leakage, and excessive agency.

Truth

Factual reliability, uncertainty handling, citation behavior, and unsupported claims.

Fairness

Political symmetry, loaded framing, refusal consistency, and decision bias signals.

Accountability

Human review, appeal paths, audit logs, disclosures, and release governance.

Ways to be evaluated

Free to start. Paid to prove it.

A Public Reality Screen is limited but real: published, inspectable artifacts and a Public Signal — useful evidence, but not a procurement-grade badge. A Full Release Verification is a deep, cooperative, release-specific review with a remediation window before anything goes public, ending in a solid Trust Grade badge the company can actually use in sales.

Payment buys the review. It never buys the grade.

See full pricing, including lower-cost options →

Free

Public Reality Screen

External, non-cooperative screen with published artifacts and a Public Signal.

From $20,000

Full Release Verification

Cooperative, release-specific audit with a remediation window and a solid Trust Grade badge. New Release Verification from $10,000 — a new dated record, never an overwrite.

Recently screened

Real projects, real records.

Public Reality Screens run against actual frontier models, external and non-cooperative, with published artifacts behind every claim.

Compare all entries on the leaderboard →

The Reality Gap

Not benchmark theater. Real, messy work.

Public benchmarks are gameable and often don't predict practical behavior — a model can top a leaderboard and still hallucinate an API or break working code the moment the task isn't a clean, self-contained problem. Algodai's headline metric is the Reality Gap Signal: how a model or product actually performs on fresh, rotating, practical tasks — real bug fixes, long documents, agentic coding with real file access — that can't be memorized in advance.

Payment buys the review. The grade is earned.

A project can pay for a Full Release Verification and still receive a weak Trust Grade. A Public Reality Screen can find real issues without any payment at all. That separation is visible on every record and verification page.