Cheapest RPC provider, priced per workload across 18 ranked providers

It depends on the workload, by multiples rather than percent. Across five workload profiles and three volumes, three providers win a cell outright: Dwellir seven, BlockPI three and Syndica two. Three more cells are ties at $49: Dwellir and Chainstack on the indexer and trace profiles at ten million, Chainstack and Helius on Solana at ten million. Read the row of the per-workload grid that matches your method mix rather than a single number, or move the calculator to your own volume.

Because they cost several times what the leaders charge for the same work. Alchemy bills $10,500 a month for a billion simple reads where Dwellir bills $1,999, and across the fifteen cells the two run between 1.7x and 10.6x the cell leader. Both are priced as premium products and both publish their rates plainly; the bench simply applies them.

They tie on recent-state reads and part ways on history. Both bill one flat unit per call, so a simple read costs 1 on either. Chainstack charges 2 Request Units for any read 127 or more blocks behind the tip and for every debug or trace call; Dwellir charges 1 credit for every response, archive and trace included. At ten million indexer or trace requests both land on $49. From a hundred million up Dwellir is cheaper and the gap widens with volume: $299 against $499 at a hundred million, $1,999 against $8,990 at a billion trace requests. Chainstack's rules were measured against its billing API; Dwellir's are its published price list. On Solana the comparison does not arise: Chainstack serves it and Dwellir does not.

By applying each provider's own method table to the same method mix, which gives weighted units per request. That figure is published per provider and per profile, so the conversion is inspectable rather than asserted. It is also why the ranking changes with the workload: the unit conversions are not constant across methods.

No. They are published prices, human-verified against each provider's own pricing page or pricing API, with the date each figure was read. No provider in this cohort was billed to confirm its rate card. Where a provider's real charge is non-deterministic, dRPC routes to third-party operators that can rewrite the price downward and exposes no cost header, the published rate is reported as an upper bound.

Because it cannot serve that workload. For the dapp mix, the calculator lists every provider that quotes nothing at the chosen volume with its reason. Infura's allowance is daily and hard-stops rather than metering overage. Alchemy's published self-serve throughput ceiling is below what a billion-request indexer workload needs. Tatum and Moralis publish 200 requests per second against the 386 that volume requires. A missing cell is a finding, not a gap in the data.

They have their own bench, 283 rpc-free-tier, ranked by what the allowance actually buys rather than by price, since every free plan costs the same nothing. Converted into simple reads per month they differ by 160x, from OnFinality's 12.0 million to GetBlock's 75,000. The allowance ranking is bench 283, rpc-free-tier. Several are limited by throughput rather than volume: GetBlock's daily allowance drains in 125 seconds at its own advertised 20 requests per second.

Every figure carries the date a human last verified it against the source, in the catalogue. The harness re-fetches 23 artifacts that carry a baseline hash and flags, in its own metrics, any whose body has changed since a human last read it. Artifacts that churn for reasons unrelated to price are re-read by hand instead. Freshness varies within a provider: Chainstack's pricing markdown has not moved since April 2026 while its pricing JSON regenerates every fifteen minutes.

Yes. The pricing catalogue, the cost model and the workload profiles are all in the harness repository linked below, and every figure carries its source URL and verification date. The weighted units per request are published as a metric so the arithmetic can be redone by hand.