Compare RPC providers

What to compare in an RPC provider

An RPC provider sits between your app and the chain. When it slows down, serves a stale block, or throttles you at peak, your users see a broken app. A useful RPC provider comparison therefore looks past a single latency number.

This page scores Chainstack, Quicknode, and Alchemy on eight pillars with published weights. Latency is only part of one pillar. Cost, Availability, data freshness, limits, compliance, and support response time cover the rest of what you feel in production.

• Reliability (25): Availability, block freshness, and data correctness or agreement; separate error-rate testing is pending.
• Cost (20): dollars per 1M real requests, not per credit.
• Performance (20): P95 latency only, averaged equally across three regions.
• Limits and RPS (10): throughput on a plan around $50, behavior at spikes, and flat-fee options.
• Coverage (5): networks, archive, debug and trace, streaming, and dedicated nodes.
• Enterprise readiness (10): SOC 2, ISO 27001, SLA with credits, and SSO.
• Support (5): declared first-response time and channels.
• AI-readiness (5): MCP server, llms.txt, machine-readable pricing, and agent skills.

Reliability uses Availability (40%), freshness (25%), and correctness or majority agreement (15%) where fresh Grafana data exists for every provider. Availability means RPC request success, not time-based uptime. The separate 20% error-rate slot remains unscored because it would double-count the same requests. Performance currently uses P95 latency alone, contributing 100% of that pillar when comparable data is available. This measures tail response latency, not complete performance or transaction-delivery speed. This is not a six-region, 30-day benchmark.

During a live refresh, Overall may display the previous comparison snapshot, explicitly marked as not current. Expired measurements are not reused in current calculations. A previous graph is displayed only while its oldest input retrieval is less than one hour old, and only for the same networks, use case and weights. Re-rendering the graph does not restart this limit.

Availability uses a logarithmic nines scale: 99% or lower scores 0, 99.99% or higher scores 100. Freshness scores the percentage of region-minute samples within two blocks (two slots on Solana) of the best-in-region tip. Freshness and proof correctness use upstream aggregates when available for every provider; otherwise, complete data for the same three regions is required for every provider and averaged equally. The evidence identifies which aggregation is used. Correctness scores 100 at zero mismatches and 0 at 0.5% mismatches, linearly; majority agreement is explicitly distinguished from proof verification. Mean regional P95 at most 80 ms scores 100; at least 1,500 ms scores 0, logarithmically. Only comparable data retrieved within ten minutes affects Overall. Missing metrics reduce scored coverage rather than becoming zero scores.

Comparing pricing and free tiers: request units, credits, and compute units

Providers bill in different units, so plan prices do not compare directly. Chainstack charges 1 request unit per request, regardless of method, and 2 RUs for archive, debug, and trace calls. Quicknode charges about 20 API credits per standard EVM call. Alchemy prices each method in compute units: eth_blockNumber costs 10, eth_getBalance 20, eth_call 26, and eth_getLogs 60.

1. List your top methods and their share of traffic.
2. Multiply each share by the provider's units per call.
3. Pick the cheapest plan that fits your volume: base fee plus overage.
4. Divide the monthly cost by requests to get dollars per 1M.

Overage rules matter too. Chainstack plans keep serving requests past quota at a per-1M overage rate when you turn on extra usage, from $15 on the Growth plan down to $5 on the Enterprise plan. Quicknode paid plans bill overage, while its Free plan stops at 10M credits. The Alchemy Free plan stops at 30M CUs. Read the unit tables in the Chainstack request units docs, the Quicknode API credits page, and the Alchemy compute unit costs before you model your bill.

What latency benchmarks miss

Latency matters, but published numbers disagree because each publisher picks its own method. In a prior capture, the Chainstack Compare dashboard (US, Germany, and Singapore, 7 days) showed Ethereum p50 and p95 of 37 ms and 84 ms for Chainstack, 39 and 89 for Quicknode, and 38 and 108 for Alchemy. On Solana, the same dashboard ranked Chainstack third, behind Quicknode and Helius.

Alchemy's own benchmark (four AWS regions, 24 hours) showed an Ethereum p95 of 14 ms for Alchemy and 120 ms for Quicknode. An independent CompareNodes snapshot (eth_getBalance, 30 minutes) put Alchemy at 29 ms and Chainstack at 36 ms.

Vendor-run tests can favor their publisher, and short snapshots vary by region and method. This scorecard uses the same Grafana panel and regions for each provider, but P95 alone is not a complete performance test. Your users also feel things a median hides:

• Stale blocks, when a node lags behind the chain head.
• Rate-limit errors at peak, and hard stops when a quota runs out.
• p99 tail latency on heavy calls such as eth_getLogs.
• Slow support when something breaks, and compliance gaps that block procurement.

Which RPC provider fits your use case

Weights change with the workload. These reference Ethereum results use the 18.69M-call monthly benchmark and exclude live measurements, ordered Chainstack, Quicknode, Alchemy. Use the interactive scorecard for current measurements.

• Production dApps (74, 55, 47): reliability and cost carry the most weight, and per-request pricing keeps bills predictable.
• DEX aggregators (76, 48, 52 in the reference comparison): quote simulation means mostly eth_call, so per-request pricing and throughput matter. Performance weighs 30. The reference comparison excludes live measurements; use the interactive charts for the current partial benchmark.
• Bridges and swap (79, 56, 54): relayers watch events across chains, so eth_getLogs volume, data correctness, and multichain coverage carry the weight.
• Trading and bots: performance weighs 40 and currently uses P95 latency alone. Treat the ranking as provisional and test from your own region.
• Wallets (72, 49, 45): reliability and cost dominate. Look at archive costs if you show transaction history.
• Indexers and analytics (74, 45, 48): cost and RPS limits dominate. Flat request-unit pricing and flat-fee RPS tiers suit heavy archive reads.
• Enterprise and fintech (71, 62, 45): SOC 2, SLA, and SSO carry 25 points of weight. Quicknode scores highest on the enterprise pillar alone (93, versus 88 for Chainstack and 48 for Alchemy).
• AI agents (75, 60, 62): MCP, llms.txt, machine-readable pricing, and agent payments carry 25 points. All three offer an official MCP server. Quicknode and Alchemy accept x402 agent payments, and Chainstack has not published support for them.

Chainstack scores lowest on support in this scorecard. Its declared response time on paid plans is under 24 hours on weekdays, and the faster 24×7 tiers are paid add-ons.

Alchemy and Quicknode alternatives

Teams look for Alchemy alternatives or Quicknode alternatives for a few recurring reasons:

• Costs grow faster than traffic because methods carry multipliers, from 10 to 60 CUs at Alchemy and about 20 credits at Quicknode.
• Free plans stop at a hard cap: 30M CUs on Alchemy and 10M credits on Quicknode.
• Throughput caps, such as 50 RPS on Quicknode Build, and no flat-fee way to raise them.
• Chain or feature gaps, such as Solana gRPC pricing, archive and trace gating, or streaming on a specific chain.
• Procurement requirements such as ISO 27001 or a published SLA percentage.

As an alternative, Chainstack bills 1 RU for standard Global Node requests and 2 RUs for archive, debug, or trace requests; Growth includes 250 RPS. It offers Unlimited Node tiers from $149/mo, 44 self-serve Global Node mainnets and 93 listed protocols across access models, Dedicated Nodes from the Pro plan, and MEV protection on Ethereum and BNB Smart Chain. Read the switch from Quicknode and switch from Alchemy guides, or follow the AI-assisted migration runbook in the Chainstack MCP server.

Another provider may fit better in some cases. Quicknode lists 77 catalogue chains across access models and a 99.99% enterprise SLA. Alchemy includes free MEV protection on Ethereum, Arbitrum, BNB Smart Chain, Base, and Solana, and publishes strong latency figures. For Solana-only workloads, compare Helius too, and see the dRPC, Helius, and Ankr switch guides. More guides: best Ethereum RPC providers and top Solana RPC providers for enterprise and fintech.

How this comparison works

• Weights and thresholds are published and fixed for a year, so adding a provider never changes anyone else's score.
• Data comes from public sources: pricing pages, docs, and trust and status pages. Anything a provider does not publish scores 0 and is marked not disclosed.
• A gap under 3 points shows as a tie.
• Chainstack publishes this page and is one of the providers compared, so the scorecard shows where Chainstack trails. Send corrections to [email protected], and we log every change.
• Prices and support response times update monthly. Enterprise and AI-readiness checks update quarterly. Read how the Compare dashboard measures latency.

Networks and multichain scoring

• Each network is scored on its own: archive, debug and trace, WebSocket, streaming, dedicated nodes, regions, MEV protection, and testnets. When you select several networks, each pillar and Overall is the equal-weighted average of its available per-network scores. Overall is normalized within each network before averaging; coverage is reported separately across all selected networks. Unavailable network scores are excluded, not scored as zero.
• Evidence-row weights are averaged across all selected networks; a row that does not apply to a network contributes no weight for that network. Evidence-row scores are weighted by their per-network row weights. Displayed percentages are rounded.
• Solana: debug and trace do not apply and are left out of the coverage checklist.
• Hyperliquid: scored on HyperEVM JSON-RPC. HyperCore /info and order-book streaming are outside the method baskets. Chainstack serves 46 of 76 /info methods.
• Robinhood Chain: mainnet launched in July 2026. Alchemy states its mainnet in an undated blog post.
• Arc: mainnet launched on Sep 16, 2026. Per-chain cost and feature detail is thin for all three providers.