RPC performance, latency, reliability, SLA, uptime, regional routing and failover claims

Comparison updated: 2026-11-04

Provider universe: 15 tracked providers. Current direct evidence coverage: 10/15 have at least one topic dimension evidenced in the latest monitored public surface; 1/15 has all four dimensions explicitly evidenced. Previously verified but currently inaccessible evidence is labeled separately rather than silently discarded.

> Evidence rule: “Not verified in monitored public evidence” means the available direct sources do not support the cell. It does not mean the capability is absent. Marketing latency and uptime figures are vendor claims, not independent benchmarks. Verify chain, region, percentile, measurement window, exclusions and contractual terms before procurement.

ProviderLatency / throughput claimUptime / contractual SLARegional routingFailover / redundancyDirect sourcesEvidence coverage
QuickNodePublicly evidenced: low-latency RPC; 50,000 RPS dedicated-cluster burst; vendor says Core RPC is 2.5× faster than competitors.Publicly evidenced: Enterprise states 99.99% contractual uptime, production-endpoint measurement and service credits.Not verified in monitored public evidence.Not verified in monitored public evidence.Home, Enterprise, Pricing2/4
AlchemyPublicly evidenced: home markets purpose-built low-latency Solana RPC and gRPC. No comparable percentile/methodology surfaced in this capture.Not verified in monitored public evidence.Not verified in monitored public evidence.Not verified in monitored public evidence.Home1/4
InfuraPublicly evidenced: networks page states “10x faster request response times”; pricing publishes credit-throughput allowances. Scope/methodology requires verification.Publicly evidenced, limited: Enterprise lists an enhanced-support SLA, but no public uptime percentage or service-credit formula was captured.Not verified in monitored public evidence.Not verified in monitored public evidence.Networks, Pricing2/4
ChainstackPublicly evidenced: GEN 2.0 promotion claims lower latency; Global Nodes are marketed for high performance.Publicly evidenced, non-contractual in captured copy: Global Nodes are described as ultra-reliable. Exact uptime/SLA terms were not captured.Publicly evidenced: Global Nodes are described as geo-balanced.Not verified in monitored public evidence.Home3/4
AnkrPublicly evidenced: 56 ms average response time; Enterprise states customizable throughput above 25,000 RPS.Publicly evidenced: 99.99% uptime claim. Contractual scope/service credits were not captured.Publicly evidenced: Enterprise states geo-distribution across 24 metros.Not verified in monitored public evidence.Home, Enterprise3/4
MoralisPublicly evidenced, qualitative: docs describe the data platform as built for low latency and reliability.Not verified in monitored public evidence.Not verified in monitored public evidence.Not verified in monitored public evidence.Docs1/4
NOWNodesPublicly evidenced: approximately 0.2-second response time and more than 15,000 transactions per second are advertised.Publicly evidenced: greater than 99.95% API uptime claim; contractual scope not captured.Not verified in monitored public evidence.Not verified in monitored public evidence.Home2/4
dRPCPublicly evidenced: low-latency positioning and plan RPS values.Not verified in monitored public evidence.Publicly evidenced: eight geo-distributed clusters.Publicly evidenced: AI load balancing is named; exact health-check and failover policy should be verified.Home, Pricing3/4
Blast APINot applicable: Blast products are deprecated.Not applicable: Blast products are deprecated.Not applicable: Blast products are deprecated.Not applicable: Blast products are deprecated; migration points to Alchemy.Deprecation noticeService inactive
BlockdaemonNot verified in monitored public evidence.Not verified in monitored public evidence.Not verified in monitored public evidence.Not verified in monitored public evidence.Home, Docs0/4
TatumNot verified in monitored public evidence.Not verified in monitored public evidence.Publicly evidenced, limited: “Intelligent RPC Routing” is named, but regions and routing policy were not captured.Not verified in monitored public evidence.Home1/4
Pocket Network / GroveNot verified in monitored public evidence for a current managed RPC product.Not verified in monitored public evidence for a current managed RPC product.Not verified in monitored public evidence.Not verified in monitored public evidence. Grove is winding down; Pocket Network continues, but today's monitored Pocket page focuses on agent data services.Grove wind-down, Pocket Network0/4 current RPC dimensions
LlamaRPCNot verified in monitored public evidence.Not verified in monitored public evidence.Not verified in monitored public evidence.Not verified in monitored public evidence.No reachable direct source on 2026-11-040/4
AllnodesPreviously evidenced; not refreshed: monitored pages were HTTP 403 on 2026-11-04.Previously evidenced; not refreshed: monitored pages were HTTP 403 on 2026-11-04.Not verified in monitored public evidence.Previously evidenced; not refreshed: monitored pages were HTTP 403 on 2026-11-04.Home, Pricing0/4 current; 3 historical
Lava NetworkNot verified in monitored public evidence.Publicly evidenced: 99.99% uptime claim. Contractual scope/service credits were not captured.Not verified in monitored public evidence.Publicly evidenced: requests are scored and routed in real time with automatic failover across independent providers.Home2/4

Cross-provider readout

  • Breadth: 10 of 15 tracked providers have at least one current, direct, topic-specific public signal in the monitored surface. Blast is an inactive/deprecated service rather than an evidence gap.
  • Completeness: QuickNode is the only provider with explicit current evidence across both performance and a quantified contractual uptime dimension, but it does not expose regional-routing or failover evidence in the captured pages. No row has unequivocal, current evidence for all four dimensions under strict semantics; dRPC's load-balancing evidence does not disclose enough policy detail to equate it automatically with contractual failover.
  • Most evidenced dimensions: latency/throughput marketing and uptime/reliability language.
  • Least evidenced dimensions: routing topology and operational failover policy. Geo-distribution, load balancing and automatic failover are related but not interchangeable.
  • Service-state changes that affect comparison: Blast is deprecated; Grove is winding down while Pocket Network continues under separate stewardship.
  • Open evidence gaps: Blockdaemon and LlamaRPC have no current verified cell; Pocket/Grove has no current managed-RPC performance cell; Allnodes could not be refreshed because monitored pages returned 403.

Buyer guidance

  1. Demand workload-specific measurements. Ask for p50/p95/p99 latency by chain, method, region and transport, plus error-rate and stale-head definitions. An average response time is not a tail-latency guarantee.
  2. Read the SLA, not the homepage. Confirm uptime denominator, endpoint scope, maintenance/exclusion policy, incident attribution, claim procedure and service-credit ceiling.
  3. Separate four architecture questions: geographic presence, request routing, node redundancy and cross-provider failover. Evidence for one does not prove the others.
  4. Test shared and dedicated products independently. Dedicated-cluster capacity and SLA terms may not apply to shared API plans.
  5. Validate benchmark comparability. Vendor “Nx faster” statements require the benchmark period, regions, chains, methods, warm/cold cache treatment and competitor selection.
  6. Run failover drills. Before procurement, test regional loss, stale nodes, rate limiting, WebSocket disconnects and provider-wide incidents against recovery-time and data-consistency requirements.

GetBlock positioning implications

  • Publish reproducible, chain/method/region-specific latency percentiles and error rates rather than only a global average.
  • Expose SLA measurement scope and service-credit language publicly; QuickNode's production-endpoint and service-credit wording is a useful transparency benchmark.
  • Describe GetBlock's regional routing, health scoring and failover behavior separately, with diagrams and incident-test evidence where supportable.
  • Distinguish shared endpoints, dedicated gateways, dedicated nodes and HA clusters in both reliability claims and pricing.
  • Avoid generic archive/gRPC parity claims: verify each chain, protocol surface and workload before comparison.

Direct-source ledger and freshness

  • QuickNode: home, enterprise, pricing — checked 2026-11-04.
  • Alchemy: home — checked 2026-11-04; changed captures did not establish a decision-relevant product change.
  • Infura: networks, pricing — checked 2026-11-04.
  • Chainstack: home — checked 2026-11-04.
  • Ankr: home, enterprise — checked 2026-11-04.
  • Moralis: docs — checked 2026-11-04.
  • NOWNodes: home — checked 2026-11-04.
  • dRPC: home, pricing — checked 2026-11-04.
  • Blast API: deprecation notice — checked 2026-11-04.
  • Blockdaemon: home, docs — checked 2026-11-04; no positive performance/reliability cell classified.
  • Tatum: home — checked 2026-11-04.
  • Pocket Network / Grove: Grove, Pocket Network — checked 2026-11-04.
  • LlamaRPC: no reachable monitored direct source on 2026-11-04 (HTTP 526/DNS failures).
  • Allnodes: home, pricing — attempted 2026-11-04; HTTP 403, so prior evidence is not presented as freshly verified.
  • Lava Network: home — checked 2026-11-04.

Next research priorities

  1. Obtain current contractual SLA/service-credit terms for Alchemy, Infura, Chainstack, Ankr, NOWNodes, dRPC, Moralis, Blockdaemon, Tatum and Lava.
  2. Find first-party architecture documentation for regional routing and failover, with special attention to health checks, stale-node detection and WebSocket behavior.
  3. Recheck Allnodes through an approved browser/manual route and locate current LlamaRPC service/status evidence without interpreting collection failures as product claims.
  4. Resolve the current Pocket Network RPC gateway/operator surface separately from Grove's wind-down notice.
  5. Collect comparable latency evidence using one documented chain/method/region/percentile model.