Nansen vs Arkham

Two price lists pointed in opposite directions

Both products sell the same scarce thing — a claim that a given address belongs to an exchange, a fund or a person — and both sell it twice, once in a browser and once over an API. On the API side, which is the only side either vendor prices in public, they meter it in opposite directions.

Nansen prices the name cheap and the label dear. Searching for an entity by name costs nothing in credits; asking for that entity's current balances costs one, or one cent by the call. Going the other way — here is an address, whose is it — is the labels endpoint, and it costs 100 credits for common labels and 500 for premium ones, one address per request. Premium is where Nansen's own cohorts live: Smart Money, alpha traders, public-figure tags. Since July 2026 the holders and leaderboard endpoints follow the same rule — passing premium_labels=true costs 150 credits a call instead of 5, and a free account gets a 403 for asking.

Arkham prices the lookup at one credit. Its card records one credit for a single address lookup, 30 for a search, 250 to 1,000 for a batch of up to 1,000 addresses, and two credits per row returned on the transfer and swap endpoints. Arkham's figures here are the ones on its card, read on 19 September 2026 and not re-read for this page.

Converted into the reader's own job

Neither vendor publishes what a credit costs in dollars on a subscription — Nansen sells extra credits "at discounted rates" with no figure attached, and Arkham quotes its API plans per customer from 100 dollars a month. So convert through the two numbers that are public.

Ten lookups a week. Nansen's Pro plan, 69 dollars a month or 49 on the annual tab, tops its included balance back up to 2,000 credits on the first of each month. That is twenty common-labels lookups or four premium ones, if you spend the credits on nothing else — roughly 3.45 dollars a common label and 17 dollars a premium one on the monthly price. Arkham's pay-per-call rate is 0.20 dollars a credit, so the same lookup is 20 cents with no account at all.

A thousand addresses from an export. On Nansen that is a thousand requests and 100,000 credits for the common set — fifty months of Pro's included floor, with the rest at a price you have to ask for. On Arkham it is one batch call of at most 1,000 credits, which fits inside the 100,000-credit trial with room to spare. The trial has its own ceiling — 10,000 unique labelled addresses per period — and the heavy endpoints are rate-limited to one request a second, which is why the batch exists.

An agent paying its own way. Both vendors accept per-request payment in USDC with no key, and the price of the thing you came for is the difference. Nansen's rails sell most endpoints at one or five cents and exclude both labels endpoints by name — "Pro subscription only (proprietary entity data)" — so an agent can buy Nansen's flows by the call but not the names on them. Arkham's x402 sells the lookup. Nansen does have the first-party MCP server and Arkham does not, but a labels call through that server draws the same 100 or 500 credits.

They are not selling the same label

The arithmetic above decides the job "name these addresses", and it decides it for Arkham. It does not decide the job most people buy Nansen for, because Nansen's expensive label is a different claim.

An Arkham label says who owns a wallet, with a confidence grade attached. Nansen's premium labels say, among other things, that a wallet belongs to a cohort Nansen has defined by performance — minimum win rates, required ROI, realised-profit floors — and the product is built to be read through that cohort: token screens filtered by who is buying, flows split by holder type, leaderboards. That is a screener over Nansen's own judgement of which wallets are worth watching, and nothing on Arkham's card is a substitute for it.

In the browser, the comparison cannot be finished from outside either app. Arkham publishes no price for any platform tier, and its entity pages read without an account. Nansen's own marketing says a free account sees labelled top holders; the in-app plan grid that would say which labels a free account sees sits behind a bot checkpoint and has not been read.

What neither of them gives you

A label with a date on it. Nansen's labels endpoint returns the current set only, with no date-range parameter and nothing in the object to say when a label was applied; Nansen lists label-history updates among the reasons a past result can change when re-run. Arkham says attribution is corrected in rare cases and publishes no correction log. Neither can answer "what was this address called on the day of the trade", which is the question a backtest or an incident review actually asks.

Evidence. Arkham's 98 and 80 per cent grades are the more honest presentation, and the badge that separates them does not survive being pasted into a chat. Nansen offers no grade at all.

A licence to publish. Nansen prohibits redistribution of labels and Smart Money data under any circumstances; Arkham's terms prohibit redistribution and derivative works of its content. Being cheap per lookup does not make Arkham's labels cheap to republish.

Which one, and when

Take Arkham when the question starts with an address and ends with an owner — a deposit list, a counterparty export, an agent that needs to know whose wallet it is looking at. It is the only one of the two that sells that lookup by the call, it batches, and at Nansen's prices the same job runs to months of a subscription.

Take Nansen when the question starts with a token or a market and ends with a cohort — which of the wallets Nansen rates are buying, and how much. That is what the 69 dollars buys, and the cheap end of its API is the data around a name rather than the name for an address. Budget label lookups separately and sparingly, because twenty a month is the whole allowance.

Take neither as evidence about a person. Both cards say the attribution cannot be audited from outside, and naming somebody on either vendor's say-so is a claim you cannot show your working for. If the answer has to be defensible, build it in Dune or Allium and own the method.