Benchmark methodology and independent proof: transparent latency tests, third-party benchmarks and reproducible methodology

Last consolidated: 2026-10-28

Comparison date: 2026-10-28

Coverage: 1/15 providers has at least one publicly evidenced dimension; 1/15 publishes benchmark results and a substantive test methodology; 0/15 has independent proof or a fully reproducible artifact bundle verified in this pass.

> Evidence semantics: “Publicly evidenced” means a direct source explicitly supports the cell. “Not verified in monitored public evidence” means the monitored/direct sources were insufficient; it is not a claim that the provider lacks the capability. A vendor-run comparison is not independent proof. A described method is not fully reproducible unless buyers can obtain enough inputs, endpoint selection rules, measurements or code to repeat/audit it.

Normalized provider matrix

ProviderPublished benchmark resultsTransparent methodologyIndependent / third-party proofReproducible artifacts or raw dataDirect sourcesEvidence coverage / freshness
QuickNodePublicly evidenced: live chain/region/window views report P50, average, P95, success rate, freshness and provider comparisonsPublicly evidenced: 15 synthetic hosts, 3 clouds, 5 regions; fresh connections; identical payload/chain/region/schedule; all attempts recorded; failures, timeouts and rate limits retained; latency uses successful responsesNot verified in monitored public evidence; benchmark is vendor-operatedNot verified in monitored public evidence; interactive history and methodology are public, but downloadable raw observations, probe code, endpoint-selection rules and rerun package were not verifiedLive benchmarks2/4; checked 2026-10-28
AlchemyNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidence—0/4; monitored 2026-10-28
InfuraNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidence—0/4; monitored 2026-10-28
ChainstackNot verified in monitored public evidence; the monitored site links a Performance Dashboard, but today’s captured pages did not expose results suitable for classificationNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceHome0/4; monitored 2026-10-28
AnkrNot verified in monitored public evidence; “56 ms avg. response time” is a first-party headline metric, not a disclosed benchmarkNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceHome0/4; monitored 2026-10-28
MoralisNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidence—0/4; monitored 2026-10-28
NOWNodesNot verified in monitored public evidence; response-time figures on the homepage lack a benchmark method in today’s evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceHome0/4; monitored 2026-10-28
dRPCNot verified in monitored public evidence; generic “low latency” positioning is not a benchmark resultNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceHome0/4; monitored 2026-10-28
Blast APINot applicable to active procurement: vendor says Blast products are deprecatedNot applicable to active procurementNot applicable to active procurementNot applicable to active procurementDeprecation noticeDeprecated; checked 2026-10-28
BlockdaemonNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidence—0/4; monitored 2026-10-28
TatumNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidence—0/4; monitored 2026-10-28
Pocket Network / GroveNot applicable to the former Grove Portal: Grove says it is winding down; current Pocket Network material monitored today is positioned around agent data services, not a comparable RPC benchmarkNot verified for current Pocket NetworkNot verified for current Pocket NetworkNot verified for current Pocket NetworkGrove farewell, Pocket NetworkTransition limitation; checked 2026-10-28
LlamaRPCNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidence—0/4; site returned 526 and docs/status DNS failed on 2026-10-28
AllnodesNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidence—0/4; all monitored surfaces returned 403 on 2026-10-28
Lava NetworkNot verified as public result data; homepage says provider performance is “benchmarked and auditable,” but today’s monitored evidence did not expose a result setNot verified in monitored public evidenceNot verified in monitored public evidenceNot verified in monitored public evidenceHome0/4; claim checked 2026-10-28

Cross-provider readout

  • Breadth: 1 of 15 providers has positive public evidence in at least one normalized dimension. QuickNode is the only provider in this pass with both visible comparative results and a substantive public methodology.
  • Completeness: 0 of 15 providers has evidence across all four dimensions. No provider was credited with independent proof, because vendor-operated tests are not independent even when competitors are included.
  • Most evidenced dimension: published benchmark results (1/15). Least evidenced: independent proof and reproducible artifacts (0/15 each).
  • Important distinction: homepage latency/uptime numbers, “fast” claims, and links named “Performance Dashboard” do not become benchmark evidence without accessible results and test conditions.
  • Lifecycle caveats: Blast API is deprecated; Grove is winding down; LlamaRPC and Allnodes could not be evaluated from their monitored public surfaces today.

Buyer guidance

  1. Ask for method, chain/network, RPC methods, endpoint tier, region, concurrency, connection reuse, timeout policy, retry policy, success denominator, sample count and observation window.
  2. Demand P50/P95/P99 plus failure/rate-limit rates and block-height/data-freshness checks; average latency alone hides tail and stale-read risk.
  3. Treat a vendor comparison as first-party evidence. Require raw timestamped observations, probe code or an independent test before using “fastest” in procurement.
  4. Verify whether public endpoints, paid shared endpoints and dedicated clusters are compared like-for-like; endpoint selection can dominate results.
  5. Re-check volatile benchmark values before sales use. This brief validates disclosure, not the accuracy of every live datapoint.

Direct sources

  • QuickNode: RPC Provider Benchmarks: Live Latency by Chain — checked 2026-10-28. Direct vendor source for live comparisons and disclosed test design.
  • QuickNode: Home and Enterprise — checked 2026-10-28. Used only to contextualize the vendor’s “2.5x faster” claim; the benchmark page is the evidence source.
  • Chainstack: Home — checked 2026-10-28. Confirms a Performance Dashboard link exists; no benchmark cell was credited from the captured material.
  • Ankr: Home — checked 2026-10-28. Headline average response-time claim only; no methodology credit.
  • NOWNodes: Home — checked 2026-10-28. Headline response-time claims only; no methodology credit.
  • dRPC: Home — checked 2026-10-28. Generic low-latency positioning only; prior weak classification was removed.
  • Blast API: Deprecation notice — checked 2026-10-28.
  • Pocket Network / Grove: Grove farewell and Pocket Network — checked 2026-10-28.
  • Lava Network: Home — checked 2026-10-28. “Benchmarked and auditable” claim without a verified public result set in this pass.

Research priorities

  1. QuickNode: locate downloadable observations, API/export, probe source, endpoint-selection policy and exact sampling cadence; test whether an independent party can reproduce a slice.
  2. Chainstack: directly inspect the linked Performance Dashboard and capture its methods, providers, windows, raw-data access and ownership.
  3. Alchemy, Infura, Ankr, Moralis, NOWNodes, dRPC, Blockdaemon, Tatum and Lava Network: search first-party benchmark/reliability papers beyond generic home/pricing pages; keep cells unverified until direct evidence is found.
  4. LlamaRPC and Allnodes: retry via alternate public documentation or manually accessible first-party surfaces because automated access failed.
  5. Seek an independent RPC benchmark source with disclosed methodology; do not use the monitored CompareNodes SOC/ISO directory as benchmark proof because that page’s scope is compliance, not latency methodology.
  6. For future GetBlock comparative content, publish a versioned harness, endpoint-tier rules, raw CSV/JSON and failure-inclusive results to occupy the market’s largest evidence gap.