Best RPC Providers for High-Throughput Workloads: 10 Providers Ranked by Sustained RPS, Burst and Price per 1,000 RPS (2026)

For sustained high throughput on a self-serve plan, Dwellir's Scale plan publishes the highest ceiling: 5,000 responses per second with 10,000 burst for $999 a month, with flat-rate Unlimited Nodes up to 1,000 RPS and dedicated clusters above that. BlockPI follows at 2,500 RPS for $299, Ankr at roughly 1,500 per endpoint, and Alchemy at about 1,150 eth_call per second once its 30,000 compute units per second are converted. The other six providers in this list stop between 500 and 1,000 RPS before you have to talk to sales.

A high-throughput workload is one where the rate is the constraint and the monthly volume is not. An indexer backfilling six months of logs, a liquidation bot polling every block, a wallet app whose traffic triples when a token launches: all three can sit well inside a monthly quota and still get rejected with HTTP 429 in the seconds that matter. For those workloads a provider's headline RPS is only the start of the question. The unit it is measured in, whether a burst is allowed, whether the cap is per account or per endpoint, and what it costs to buy more all decide whether the number on the pricing page is the number you get.

This post ranks 10 providers on those criteria. Every figure was checked against the provider's own pricing page or rate-limit documentation on 5 November 2026. For the full unit conversion maths behind the RPS figures, see which RPC provider has the highest RPS; for how flat-rate products work, see what an unlimited RPC node is.

High-Throughput RPC Providers Compared (2026)

RPS figures are eth_call-equivalent where the provider meters in compute units or credits. Helius is omitted from the ranking because it serves Solana only; it is covered in the decision guide below.

The ranking splits into three bands. Dwellir, BlockPI, Ankr and Alchemy clear 1,000 RPS on a self-serve plan. thirdweb, GetBlock, dRPC and Chainstack sit between 600 and 1,000, and each sells a flat-rate or dedicated step up. QuickNode and Infura stop at 500, and QuickNode's three higher tiers add credits without adding rate. Within each band, the columns that move a provider up are a published burst, a per-response unit and a visible price for the next step.

How the Providers Are Ranked

Five criteria, in order of weight. The rate limits comparison explains the mechanics behind each one.

1. Sustained rate at the top self-serve tier, converted to eth_call per second where the provider meters in compute units or credits.
2. A published burst figure. Only Dwellir (bucket depth) and Alchemy (10-second window) document one.
3. Limit scope. Per-endpoint limits let you buy capacity by adding endpoints; per-account limits do not; per-IP limits penalise anything behind a NAT gateway.
4. A flat-rate product, which turns the request meter off and sells a rate. See what an unlimited RPC node is for how the four offers compare per RPS.
5. A published path above the shared ceiling, meaning a dedicated node or cluster with a visible starting price.

1. Dwellir

Dwellir publishes the highest self-serve ceiling in this comparison and is one of only two providers that publishes a burst figure. The rate limit is counted in responses: one method call, one WebSocket message or one gRPC stream message is one response, regardless of how heavy the method is. A debug_traceTransaction and an eth_blockNumber draw down the same bucket at the same rate.

Throughput facts:

• Scale, $999/month: 5,000 responses per second sustained, 10,000 burst, 500M responses included, $2 per million overage
• Growth, $299/month: 500 sustained, 2,500 burst. Developer, $49/month: 100 sustained, 500 burst. Free: 20, no burst
• Token bucket with the bucket depth equal to the burst figure, refilling at the sustained rate, documented on the rate limits page
• Limit applies per account. Extra keys divide the budget rather than adding to it
• Each call in a JSON-RPC batch counts as one response; no maximum batch size is published
• Unlimited Node: 25 RPS for $119, 100 for $479, 250 for $1,199, 500 for $2,399, 1,000 for $4,799 per month, per endpoint, with request metering off
• Dedicated clusters with no rate limit, from $2,400/month for an Ink archive cluster to $12,600/month for Ethereum, other chains quoted
• 99.99% uptime SLA, bare-metal infrastructure across North America, Europe, Asia and Africa, 170+ networks

Considerations: The free tier's 20 RPS with no burst is the smallest in this list for prototyping a bursty client. The Unlimited Node is served from shared capacity, so its RPS is an upper limit rather than a guarantee, which the product page states. Above 1,000 RPS flat-rate, the choice is the metered Scale plan at 5,000 or a dedicated cluster.

Best for: Workloads that burst. The 10,000-deep bucket on Scale absorbs a traffic spike that would return 429s on a strict per-second gateway, and the per-response unit means a trace-heavy or log-heavy mix gets the full rate. Teams that want a known invoice for an indexer should look at the Unlimited Node rungs; teams that need throughput as a contract should price a cluster.

2. BlockPI

BlockPI sells the most RPS per dollar on sticker price, with a catch in the billing model. Plans are packages of request units that expire, not monthly subscriptions, and each package carries both an RPS cap and a request-unit-per-second cap.

Throughput facts:

• Premium, $299: 2,500 RPS and 100,000 RU/s, 4,000M RU included, expires after 90 days
• Elementary, $49: 1,000 RPS and 40,000 RU/s, 500M RU, expires after 60 days
• Free: 20 RPS, 50M RU, 32 days. Pay as you go: 20 RPS
• eth_call costs 20 RU, so the RU/s ceiling on Premium would allow 5,000 eth_call per second; the 2,500 RPS cap binds first. Archive calls add 30% RU
• Limits apply per account. If you hold several packages, the highest package's rate applies to all of them
• Batch: up to 1,000 calls on private HTTPS endpoints, 10 on public. Private WebSocket endpoints do not support batches
• No published burst, no published SLA, no dedicated product. Enterprise offers "no hard cap on QPS" through sales

Considerations: Because the package expires, the $299 Premium costs about $100 a month if you use it for the full 90 days, and more if the 4,000M RU run out first. A log-heavy workload hits the RU/s ceiling before the RPS ceiling. With no burst figure, treat 2,500 as a strict per-second limit until you have load-tested.

Best for: Steady, predictable high-rate workloads on light methods, such as a block-following service or a balance poller, where 2,500 RPS of eth_call-class requests at $100 a month is hard to beat.

3. Ankr

Ankr prices by usage rather than by plan, and its Premium tier applies the rate limit per endpoint, which makes it easy to isolate an indexer from a user-facing API by creating two endpoints.

Throughput facts:

• Premium: approximately 1,500 requests per second per endpoint on EVM chains, around 4,000 on Solana. Ankr describes this as a load-dependent approximation rather than a fixed cap
• Pay as you go: $0.10 per million credits, 200 credits per EVM request, so $20 per million requests. Subscription deals from $500 to $3,000 a month add 20% extra credits
• Batch: 1,000 calls on Premium, 100 on Freemium, 10 on public endpoints
• WebSocket on Premium only, 500 concurrent connections per source IP across HTTPS and WS, 15 MB maximum message
• gRPC available at 10 credits per call plus 5M credits per GB
• No published burst, no SLA, no flat-rate product. Enterprise is custom

Considerations: Ankr describes the 1,500 figure as load-dependent, and there is no burst. Usage pricing at $20 per million means a 1,000 RPS workload running flat out costs about $52,000 a month, so the per-endpoint cap is really for bursts above a modest average, not for sustained saturation.

Best for: Multi-service deployments that need separate budgets per endpoint, and Solana workloads that need more than the EVM ceiling.

4. Alchemy

Alchemy is the second provider that documents its burst behaviour, and the only one in this list where you can buy rate in increments without changing plan. The limit is in compute units per second and applies per account.

Throughput facts:

• Pay as you go includes 10,000 CU/s. Additional 5,000 CU/s blocks cost $160 a month each, up to a self-serve ceiling of 30,000 CU/s
• eth_call is 26 CU, so 10,000 CU/s is about 385 eth_call per second and 30,000 CU/s is about 1,150. eth_getLogs at 60 CU gets roughly half that
• Usage billed at $0.525 per million CU in the docs ($0.52 on the pricing page)
• Throughput is evaluated over a 10-second rolling window, so a 10,000 CU/s account can spend 100,000 CU inside any 10 seconds. Alchemy also says the cap is elastic upward when capacity allows
• Batch: 1,000 calls over HTTP, 20 over WebSocket. Trace, debug and several enhanced APIs cannot be batched
• Enterprise dedicated throughput "starting from 1,000 requests per second", custom SLA

Considerations: The CU unit means the same plan gives a trace-heavy workload a fraction of the eth_call figure. Reaching the 30,000 CU/s ceiling costs $640 a month in rate blocks before any usage, so the effective price per 1,000 eth_call/s is about $555 plus consumption. The per-account scope means every app on the account shares the budget.

Best for: Teams already on Alchemy's enhanced APIs whose traffic is read-heavy and bursty rather than log-heavy, and who value the elastic 10-second window over a hard bucket.

5. thirdweb

thirdweb publishes one number per plan and has moved to "request units per second" wording, with no public conversion table.

Throughput facts:

• Scale, $499/month: 1,000 request units per second. Growth, $99/month: 250. Pro, $1,499/month and up: "No Rate Limits"
• 1M requests included, then $8 per million up to 250M, $6.80 up to 1B, $5.40 above
• Infrastructure SLA 99% on Growth, 99.9% on Scale
• Burst, limit scope, batch limits and dedicated options are not documented

Considerations: The Pro tier is the cheapest published "no rate limits" plan in this list, but it is a shared-infrastructure promise without a burst or capacity figure attached. The overage rate sits in the upper half of this list, behind Ankr and Chainstack, so a sustained 1,000 RPS workload would cost about $16,000 a month in usage on the tiered rates.

Best for: Teams on thirdweb's wallet and contract SDKs whose RPC traffic is modest on average and needs a high ceiling for launches.

6. GetBlock

GetBlock raised its shared RPS ceilings in July 2026 and sells a per-chain flat-rate node alongside them.

Throughput facts:

• Pro, $499/month: 800 RPS, 1B CU. Premium, $699/month: 800 RPS, 1.6B CU. The two plans differ in volume, not rate
• Limitless Node, per chain, request metering off: EVM full node 25 RPS for $150, 50 for $300, 150 for $500, 300 for $1,000. EVM archive 15 to 100 RPS for $150 to $1,000. Solana 5 to 50 RPS for $150 to $1,500
• Dedicated nodes from $1,000/month with unlimited requests and unlimited RPS, 99.99% uptime
• WebSocket on all tiers. EVM Data Streams: 50 connections, 15 subscriptions
• No published burst. Whether the RPS cap is per key or per account is not stated. No batch limit published

Considerations: The Limitless Node tops out at 300 RPS, below the 800 RPS shared cap, so it changes the billing model without raising the ceiling. The product page's hero says "up to 200 RPS" while the pricing table lists a 300 RPS tier; the table is the one to trust.

Best for: Mid-rate workloads that want a flat invoice for one chain, and teams that want a dedicated node with a published starting price.

7. dRPC

dRPC bills a flat 20 compute units per method, so its CU figures convert to requests at 1:20 and the rate limit behaves like plain RPS.

Throughput facts:

• Scale, $199/month: 600 RPS, 664M CU (about 33M requests at 20 CU each), $6 per million overage. Pro, $49/month: 300 RPS. Pay as you go: $6 per million at 100 RPS
• Free: 210M CU per month on public nodes, rate limited per IP at 120,000 CU per minute (about 100 eth_call/s) with a floor of 50,400 under load
• Enterprise from $990/month: unlimited RPS, dedicated clusters, committed SLAs, no figures published
• Batch limited to 3 calls on the rate-limited free tier
• No published burst. WebSocket and gRPC limits not published

Considerations: The "5K RPS" that dRPC marketed on its Growth plan earlier in 2026 is gone; the top self-serve plan is now 600 RPS and anything above that is Enterprise. The per-IP free limit catches Kubernetes deployments where every pod shares a NAT address. The dRPC pricing pages did not load for our check, so the Scale figures come from search-engine extracts and a third-party listing; verify before signing.

Best for: Cost-sensitive workloads under 600 RPS, where $199 for the rate and $6 per million is the cheapest metered combination in the list after BlockPI's packages.

8. Chainstack

Chainstack pairs a plain RPS cap with a monthly request-unit quota and sells an Unlimited Node add-on with the widest published rung range after Dwellir.

Throughput facts:

• Business, $499/month: 600 RPS, 200M RU, $10 per million RU overage. Enterprise from $990: "unlimited" RPS
• Unlimited Node add-on, requires Growth plan or higher: 25 RPS for $149, 100 for $649, 250 for $1,649, 500 for $3,199. The docs list a 1,000 RPS tier with no price
• Dedicated nodes on Pro and above from $0.50 per node-hour plus $0.01 per 20 GB-hour of storage
• WebSocket: 500 concurrent connections, 1-hour idle timeout. HTTP connections close after 1,000 requests
• No published burst. The docs call the ceiling a "global plan RPS limit", so every endpoint in the account shares it

Considerations: Chainstack's shared ceiling is the second lowest among the providers with a flat-rate product, above only QuickNode, and its Unlimited Node costs $6,398 per 1,000 RPS on the 500 rung, second only to QuickNode's Flat Rate RPS at $7,796 on the 250 rung. The dedicated node hourly price is the most transparent in the comparison for teams that want their own hardware.

Best for: Teams that want a dedicated node priced by the hour, and multi-chain teams already on Chainstack who need a predictable invoice for one hot endpoint.

9. QuickNode

QuickNode runs the strictest gateway in this list: no bucket, and its docs state that 100 requests arriving in one second on a 75 RPS endpoint means 25 rejections.

Throughput facts:

• Business, $999/month: 500 RPS, 2B credits. Business+ tiers at $1,499, $1,999 and $2,999 keep 500 RPS and add credits
• Flat Rate RPS on EVM: 75 RPS for $799, 150 for $1,499, 250 for $1,949 per month, Solana at 1.5x. Above 250 RPS is sales-only; custom setups beyond 2,500 RPS
• Flat Rate RPS scope: one endpoint, one chain, one region, with 2 to 6 concurrent HTTP connections and 10 to 30 WebSocket connections by tier
• Flat Rate RPS runs on shared infrastructure with no SLA guarantees
• Shared plan limits apply per account; batch and WebSocket caps on shared plans are not documented

Considerations: $999 for 500 RPS is the highest price per 1,000 RPS in this comparison, and the next three tiers add nothing to the rate. The flat-rate product's per-endpoint connection caps are tight for a service that fans out requests.

Best for: Teams on QuickNode's add-on marketplace whose rate needs are moderate, and single-region workloads that want a named RPS figure on one endpoint without metering.

10. Infura

Infura publishes no RPS figure anywhere. Everything below is converted from credits.

Throughput facts:

• Team, $225/month: 40,000 credits per second and 75M credits per day. eth_call is 80 credits, so about 500 eth_call per second, about 157 eth_getLogs at 255 credits, or 40 debug_trace* calls at 1,000 credits
• Developer, $50/month: 4,000 credits/s, about 50 eth_call/s
• The daily credit cap is a second ceiling: 75M credits a day is about 940,000 eth_call, or 500 RPS for 31 minutes. WebSocket connections are cut when the daily limit is reached
• Every call in a batch is charged, plus the batch call itself; no size limit stated
• No burst, no flat-rate product, Enterprise through sales with committed SLAs

Considerations: The daily cap is the figure that matters for high throughput. A workload that actually uses the 500 eth_call/s ceiling exhausts the day's credits in about half an hour. The pricing page contradicts itself on the free tier (3M versus 6M credits a day), which is not reassuring for the rest of the table.

Best for: Low-average workloads that need brief headroom, and teams in the MetaMask ecosystem. Not a fit for sustained high rate without an Enterprise agreement.

Which Provider for Which High-Throughput Workload

Indexer backfill, hours of eth_getLogs at full rate. Sustained rate decides this one, and burst barely matters. Prefer per-response pricing over CU and credit metering, which charge log calls at a multiple. Dwellir Scale (5,000 responses/s, every call counted once) or a Dwellir Unlimited Node at 500 or 1,000 RPS gives a known invoice. BlockPI Premium is the budget option if its RU/s ceiling fits your range widths.

Trading or liquidation bot, low average and sharp spikes. Burst depth decides this one. Only Dwellir and Alchemy publish it. Dwellir Growth at $299 gives a 2,500-deep bucket on 500 sustained; Alchemy's 10-second window gives a similar effect on its CU budget. Add client-side queuing either way, and treat QuickNode's strict per-second gateway as a reason to buy more RPS than your average suggests.

Consumer app with launch-day traffic. Per-account scope and a burst allowance keep one spike from taking every service down. Dwellir Scale or Alchemy at the 30,000 CU/s ceiling. If you would rather isolate services, Ankr's per-endpoint 1,500 RPS or a second QuickNode Flat Rate RPS endpoint gives the public API and the indexer separate budgets.

Solana. Helius Professional at $999 gives 500 RPS with 100 RPS on its DAS and enhanced APIs, 100 sendTransaction per second and gRPC streaming on Business and above, with dedicated nodes from $2,900. Ankr's Solana ceiling of around 4,000 per endpoint is higher on paper but is a load-dependent approximation on usage pricing.

More than 5,000 RPS, or a contractual rate. No self-serve plan in this list goes above Dwellir's 5,000. Dwellir dedicated clusters start at $2,400 a month by chain, dRPC Enterprise at $990, GetBlock dedicated at $1,000, Chainstack at $0.50 per node-hour. Alchemy, QuickNode and Infura price it through sales.

Conclusion

For most high-throughput workloads, Dwellir is the best all-round choice: the highest published sustained rate at 5,000 responses per second, a documented 10,000 burst, matched in spirit only by Alchemy's 10-second window, a per-response unit that does not penalise heavy methods, and a published path from flat-rate Unlimited Nodes to dedicated clusters when shared capacity is no longer enough.

Two alternatives earn a place depending on the workload. BlockPI Premium is the cheapest way to buy 2,500 RPS for light methods if you can live with 90-day packages and an RU/s ceiling. Alchemy is the pick for teams already on its enhanced APIs whose traffic is read-heavy and bursty, because its 10-second window and $160 rate blocks let you grow throughput without a plan change. The rest of the list serves the volume end of the market well. For the rate end, check the burst, the unit and the scope before you trust the headline.

Ready to test a bursty client against a published bucket? Create a free Dwellir account, or price a sustained rate on the RPC provider pricing calculator.

Frequently Asked Questions

Which RPC provider is best for high throughput?

On published self-serve plans, Dwellir's Scale plan has the highest sustained ceiling at 5,000 responses per second with 10,000 burst for $999 a month, and it also sells flat-rate Unlimited Nodes up to 1,000 RPS and dedicated clusters above that. BlockPI publishes 2,500 RPS for $299, Ankr about 1,500 per endpoint, Alchemy about 1,150 eth_call per second at its 30,000 CU/s self-serve ceiling. Everyone else tops out between 500 and 1,000 RPS before you talk to sales.

Metered plan, unlimited node or dedicated cluster: which is right for high throughput?

A metered plan when your average is well below the ceiling and you want burst headroom: Dwellir Scale gives 5,000 sustained and 10,000 burst for $999 plus $2 per million above 500M. A flat-rate unlimited node when the rate is steady and the volume is large, such as an indexer at 500 RPS around the clock, which would cost about $2,600 a month on Scale once overage is added but $2,399 flat on a 500 RPS Unlimited Node with no metering. A dedicated cluster when you need the rate as a contract rather than a shared ceiling, or more than 5,000 RPS.

Does buying a second API key or endpoint double my RPS?

Only where the limit is scoped per endpoint. Ankr Premium and QuickNode Flat Rate RPS attach the rate to one endpoint, so a second endpoint adds capacity and a second invoice line. Dwellir, Alchemy, BlockPI, Chainstack and Infura apply the limit per account, so a second key shares the same budget. dRPC's free tier limits per IP address, so every pod behind one NAT gateway shares one allowance.

Why does my high-throughput workload get 429 errors below the published RPS?

Three reasons. The limit may be in compute units or credits, so heavy methods use it up faster than the eth_call figure suggests. The limit may be per account, so a second key or service shares the same budget. Or there is a second ceiling under the RPS cap, such as a daily credit quota (Infura), a request-unit rate (BlockPI) or per-method sub-limits (Helius).

What is the cheapest way to buy 1,000 RPS?

Per 1,000 sustained RPS at the top self-serve tier, on a monthly basis: BlockPI at about $40 ($299 per 90-day package, 2,500 RPS), Dwellir Scale at $200, dRPC at $332, Infura at $450, thirdweb at $499, Alchemy at about $555 in rate blocks plus usage, GetBlock at $624, Chainstack at $832, QuickNode at $1,998. Flat-rate products are priced differently: Dwellir's 1,000 RPS Unlimited Node is $4,799 a month with no request metering at all.

Sources

All provider figures checked on 5 November 2026.

• Dwellir rate limits, Unlimited Nodes, dedicated clusters
• BlockPI pricing and rate limits, best practices
• Ankr service plans
• Alchemy pricing, throughput, compute unit costs, batch requests
• thirdweb pricing
• GetBlock pricing, CU and rate limits, Limitless Node
• dRPC pricing, dRPC rate limit docs
• Chainstack pricing, RPS plan limits, Unlimited Node
• QuickNode pricing, Flat Rate RPS
• Infura pricing, credit costs
• Helius rate limits, pricing