Every team that ships a wallet, a DeFi dashboard or a treasury report hits the same wall: blockchain data is public, but making it queryable is not free. The vendors that solve this — Dune, Covalent's GoldRush, Nansen, The Graph and Goldsky — all cover Ethereum mainnet and Arbitrum One, and all of them meter usage. What matters at procurement time is not which platform is fastest. It is how each one counts a unit of work, and what happens to your invoice when activity shifts from mainnet to a high-throughput layer 2.
Three different jobs sold under one label
"On-chain data management" bundles work that buyers routinely conflate, and picking the wrong category is the most expensive mistake in this market.
- Analyst-facing SQL and dashboards. A hosted warehouse where humans write queries. Dune is the reference implementation.
- Application-facing REST APIs. Endpoints your product calls at runtime for balances, transfers and token metadata. GoldRush and Nansen's API sit here.
- Pipelines into infrastructure you own. Subgraphs or streams that land decoded events in your own Postgres, ClickHouse or Kafka. The Graph and Goldsky occupy this tier.
A team that buys a dashboard tool and then serves production traffic from its API pays a premium for the mismatch. The reverse holds too: building pipelines to answer one quarterly question is engineering effort nobody asked for.
How the metering models differ
Dune
Dune organizes data into three layers: raw block, transaction, log and trace data indexed from nodes; decoded contract events and function calls parsed with verified ABIs; and curated datasets maintained by its own data team. Per Dune's data catalog documentation, Ethereum and Arbitrum both get full coverage across all three, including DEX trades, token transfers, balances, prices, bridges, lending and stablecoins. Raw data lands within minutes of finality; curated tables refresh hourly.
Billing runs on credits that scale with the compute a query consumes, so a simple lookup costs less than a heavy join. Dune's pricing FAQ puts the free tier at 2,500 monthly credits on the small engine only, with a two-minute timeout, three concurrent executions, and no scheduling. API access is included on every plan, free tier included, but still draws down credits. Overage rates are where tiers separate: $5.00 per 100 extra credits on Free, $1.875 per 100 on Analyst monthly, $1.596 per 100 on Plus monthly, with annual commitments cheaper on both paid tiers. API exports meter at one credit per 1,000 datapoints on Free and Analyst, and one per 5,000 on Plus — a five-fold efficiency gain easy to miss when comparing headline prices.
Covalent GoldRush
GoldRush sells a unified multichain REST API rather than a query surface. Its published pricing lists a free trial with 25,000 credits at 4 requests per second, a $10-per-month "Vibe Coding" tier with 10,000 credits at the same 4 RPS aimed at agents and prototypes, and a $250-per-month Professional tier with 300,000 credits at 50 RPS. Overage is explicit: $0.001 per flex credit on Professional and $0.00077 on the entry tier. A custom "Inner Circle" plan adds SLAs, NDAs and dedicated support. GoldRush is also resold through Google Cloud Marketplace and QuickNode, both starting at $50 per month — relevant if you would rather draw down existing cloud commitments than add a vendor.
The binding constraint here is the rate limit, not the credit count. Four requests per second is a prototyping budget; any consumer-facing feature that fans out several calls per page load needs the Professional tier for throughput long before it needs it for volume.
Nansen
Nansen's value is labeled entity data — wallet clustering and attribution — rather than raw completeness, and its metering reflects that. Nansen's credits documentation prices Pro at $49 per month billed annually or $69 monthly, bundling 2,000 starter credits with top-ups that expire a year after purchase. The free plan grants 100 trial credits refilling at 10 per day, with access to all Pro endpoints.
Endpoint costs vary by two orders of magnitude. Foundational calls — balances, trades, transfers — cost one credit each. Mid-tier endpoints run 5 to 25. Label endpoints range from 100 to 500, and the agent endpoint costs 200 credits in fast mode or 750 in expert mode. Budgeting from the 2,000-credit allotment alone misleads badly: that allowance is either 2,000 balance checks or roughly two dozen premium label lookups, depending entirely on which endpoints your code touches.
The Graph and Goldsky
If you want data inside infrastructure you control, the economics change shape. The Graph's Subgraph Studio documentation confirms a free plan of 100,000 monthly queries, with usage beyond that documented at $4 per 100,000 queries. Notably, the protocol's billing contracts now run on Arbitrum One rather than Ethereum mainnet — you pay in GRT or by card, and gas for those billing transactions is denominated in ETH on Arbitrum. Goldsky takes the managed-vendor route to a similar outcome, offering hosted subgraphs plus Mirror pipelines that stream decoded events into Postgres, ClickHouse, S3 or Kafka, with a free Starter tier covering community subgraphs.
Why Arbitrum changes the arithmetic
Ethereum mainnet and Arbitrum One are not interchangeable line items even when a vendor supports both at parity. Two effects dominate.
First, volume. Arbitrum exists because blockspace there is cheap, so applications emit far more events per unit of economic activity than they would on mainnet. Under any credit-metered or per-query model, identical user behavior costs more to observe on the L2. Teams that size a data budget from mainnet history and then migrate are the ones who get surprised.
Second, fee accounting. Arbitrum transactions carry an L1 data-posting component alongside L2 execution cost, so "gas spent" is not one number. Dune's catalog documents dedicated gas-breakdown curated datasets for Arbitrum for exactly this reason. If your reporting depends on accurate unit economics, confirm your vendor decomposes L2 fees rather than flattening them.
What this comparison deliberately does not claim
You will find articles ranking these platforms by query latency and indexing lag. Treat those skeptically, including vendors' own figures. Sub-second indexing claims are marketing positions, not audited results, and no number in this piece is a performance benchmark — these are published prices and documented limits, verifiable in a way throughput claims are not.
Benchmark it yourself, because doing so is nearly free. Every platform here has a no-cost tier. Take one contract you care about, deployed on both Ethereum and Arbitrum, run your three real queries against each candidate for a week, and record credit burn alongside wall-clock time. That beats any third-party leaderboard.
Choosing
Pick Dune if analysts rather than services are the consumers, and if curated cross-protocol tables save you weeks of modeling. Pick GoldRush if you need one REST contract across many chains and can live inside its rate limits. Pick Nansen if labeled wallet attribution is a hard requirement — nothing else here substitutes for it, and nothing else charges 500 credits per call either. Run subgraphs or Mirror pipelines if the data must sit beside your existing warehouse.
One rule holds regardless of tier: instrument your consumption before signing an annual contract. Every vendor here meters something different — compute, requests, datapoints, queries — and the plan that looks cheapest on a pricing page is frequently not the one that survives production traffic.