Pricing and product facts were re-checked in October 2026 and drift: verify current terms directly. We compete with the service discussed, which is exactly why this piece argues from architecture rather than adjectives. Nothing here is investment advice. Decisions are yours.
Token Metrics sells AI-generated grades for thousands of tokens, from a free tier to a Premium plan near $200 a month. This site publishes seven plain measurements over one hundred assets, free. The comparison is not about polish or coverage, where Token Metrics wins, but about a single architectural question: can you check the number you are given?
What each product actually is
Token Metrics is a research platform whose centerpiece is scoring: AI-derived trader and investor grades across a vast token universe, plus indices, analytics, and editorial research, priced in October 2026 at $39.99 a month for Advanced and $199.99 for Premium, with a free Basic tier and an application-only VIP tier above them. Whether it is worth buying is a separate verdict; this piece is the narrow architectural comparison. Our board is a measurement instrument: seven published formulas over a rule-governed top-100, spelled out in the monitor methodology, plus a two-condition regime dial. No composite score, and no coverage ambition beyond the liquid market that is actually tradeable, which is what the universe rules exist to fence off.
The grade versus the formula
A grade compresses; a formula discloses. Token Metrics' scores are proprietary by design: inputs, weights, training and validation are not published, so a grade of 80 is an argument from authority, which is the general problem with AI trading claims. It may encode real insight, and you cannot distinguish that from noise, drift or marketing, because nothing about it is reproducible. Our columns are the opposite trade. They encode less, momentum, trend state, volatility, liquidity, distance from high, and every one can be recomputed by a stranger tonight. The case against secret sauce makes the philosophical bet explicit: seven checkable numbers inform better than one unfalsifiable grade, because the reader can see why, which is the whole point of reading the board column by column.
The coverage question, inverted
Thousands of tokens versus one hundred reads as a point for the platform, and for idea generation it honestly is. Invert it once for completeness. A universe that size is mostly tail: assets whose books are too thin to enter or exit at meaningful size without paying the thin-book spread, which makes a grade on them a judgment you cannot cheaply act on. The tail is also where the graveyard grows fastest, so wide historical coverage inherits the survivorship question in proportion to its width.
Breadth also multiplies the showcase problem. Grade thousands of assets and some subset will always have soared after receiving a high mark, by arithmetic alone; those become the marketing exhibits, while the misses dissolve into a universe too large for anyone to audit by hand. Our hundred-asset limit is not modesty. It is the size at which every row stays individually checkable and the whole board can be recomputed by one stranger in one evening.
The record question
Both operations claim their outputs are useful; the test is what binds the claim. Grade performance at Token Metrics is reported by Token Metrics, self-compiled and revisable, which is the industry norm and most of the 95% lie, and its public review record is notably mixed, a fact the verdict piece treats in full. This site's outputs are attested daily before outcomes arrive, and you can verify any day of it yourself. They carry no performance claim at all until the derivation gates pass in public, somewhere along the path from measurement to signal. So the honest current comparison is a checkable record of measurements against an uncheckable record of grades.
What would change this comparison
The architectural gap is not permanent, and honesty requires naming what would close it. Three publications would move the grades from unfalsifiable to checkable: point-in-time universe documentation for every historical performance figure, dead tokens included; a validation write-up with the tuning window, trial count, and out-of-sample results stated; and third-party or cryptographic attestation of grade history, so past scores cannot be revised after outcomes arrive. None of this requires disclosing the model itself. A black box with an attested output ledger is a perfectly auditable product, and the ledger, not the box, is what this comparison weighs.
The offer runs the other way too. Our architecture is criticizable precisely because it is published: if the seven columns measure the wrong things, or the dial's two conditions are naive, the formulas are there to attack, and documented criticism improves the instrument. Both products would be better under the other's scrutiny. Only one of them is currently structured to receive it, and that asymmetry, not any adjective, is the comparison's finding.
Where Token Metrics genuinely wins, and the choice
Fairness requires the other column. Token Metrics covers thousands of tokens where we deliberately cover a hundred; it ships indices, screeners, editorial research, and a mature product surface; and a free Basic tier means trying it costs nothing but attention. A reader who wants broad-universe idea generation with an AI layer, and who is comfortable trusting grades they cannot audit, is describing that product, and the money question for them belongs to that separate verdict, not to this one. A reader who wants fewer numbers they can fully verify, a regime instrument with published thresholds, and a research program whose every claim arrives pre-registered because it is built in public, is describing this site, at a price of zero. Who pays for that, and why, is the economics of free. The architectural line is the whole comparison: trust the model, or check the math. Decisions are yours.