RPC provider performance on BNB Chain: a regional latency and reliability benchmark (October 2026)

By Chainstack Labs Research Team · Research Note No. 052 · October 2026 edition · Published Oct 8, 2026 · Last updated Oct 8, 2026 · Node infrastructure

Data source: Chainstack Compare, backed by a public Grafana dashboard. Seven-day rolling window across DE, US, and SG, snapshot captured Oct 8, 2026.

Disclosure. Chainstack operates the monitoring service, the dashboard, and the scoring formula behind this report, and Chainstack is one of the four providers ranked. Section 2 documents the pipeline so readers can check the work, and the collection code is open source.

---

Abstract

BNB Chain’s throughput puts pressure on RPC providers to hold latency under load, which makes the median a weak guide on its own. This report ranks Chainstack, Quicknode, Alchemy, and dRPC on BNB Chain RPC latency and uptime over a 7-day window ending Oct 8, 2026, using the scoring method behind Chainstack Compare: response time and success rate are combined per region, and the three regions (DE, US, and SG) are averaged equally. Chainstack ranks first on score, but no provider leads every percentile.

Keywords: BNB Chain RPC provider, BSC RPC provider, BNB Chain RPC latency, RPC benchmark, regional latency, availability, composite scoring, tail latency, methodology

Key findings

• Chainstack ranks first with a score of 0.087 and the lowest P95 (76 ms).
• It does not lead every column. Alchemy has the lowest P50 (32 ms against 35 ms), and Quicknode (193 ms) and Alchemy (222 ms) both have a lower P99 than Chainstack (270 ms).
• Second place is a tie. Alchemy (0.135) and Quicknode (0.136) differ by 0.001, which is inside rounding.
• dRPC’s September outlier did not persist. Its score fell from 0.419 to 0.200 as DE P95 dropped from 559 ms to 219 ms and US from 689 ms to 248 ms.
• Chainstack’s lowest cell is SG at 29 ms, and its US P95 fell from 148 ms to 101 ms since September.

---

Contents

1. Introduction
2. Methodology and scoring
3. Providers and regions
4. Results
5. Regional variance and outliers
6. Changes since September
7. Interpretation notes and known confounders
8. Conclusion
9. Frequently asked questions
10. Appendix A: measurement parameters

---

1. Introduction

BNB Chain (also called BNB Smart Chain, or BSC) is tuned for throughput through short block times and high transaction volume, so an RPC provider serving it faces denser and burstier request patterns than on a quieter chain. A provider that looks fine at the median can behave differently in the tail, which is why this report reads the full P50, P95, and P99 spread and the regional detail instead of the headline score alone.

2. Methodology and scoring

2.1 Monitoring infrastructure

A Python service running on Vercel serverless functions sends RPC requests to each monitored provider every 1–3 minutes from three regions and records each request’s response time and outcome. The results feed a public Grafana dashboard, which also powers Chainstack Compare. Regions are labeled by country code: DE, US, and SG. Because probes originate from cloud regions, the benchmark measures latency between each provider and those regions, not the latency an end user on a home or mobile connection would see.

2.2 What is requested and what counts as a failure

All requests use the latest block, except methods marked with an hourglass (⌛) on the dashboard. These “archive” methods query historical blocks one to two days old, which exercises a provider’s historical-state path rather than its tip-of-chain path. A response that takes longer than 55 seconds is treated as a failure, not as a very slow success, and is excluded from the latency charts. Per-method breakdowns are available in Grafana; this report uses the summary view.

2.3 Plan monitored

Unless a different plan is specified, each provider is monitored on its first paid RPC plan, named Growth as a rule. The results describe each provider’s entry paid tier. Dedicated, enterprise, and other higher tiers are not covered, and rate limits or routing that differ by plan may affect both latency and success rate.

2.4 Scoring formula

The dashboard ranks providers on speed (response time) and reliability (success rate). Lower is better.

Score = 1 / ((1 / ResponseTime) × (SuccessRate³))      = ResponseTime / SuccessRate³

The score is built in four steps:

1. Latency term. For each region, take each method’s p95 and combine them as an arithmetic mean with equal weight per method. eth_subscribe is excluded from this average because its latency reflects chain block cadence, not provider speed.
2. Success-rate term. For each region, compute the success rate using sample counts as weights. eth_subscribe still counts toward this term.
3. Per-region score. Divide the latency term by the success rate cubed. Response time is in seconds, so a latency term of 0.087 means 87 ms.
4. Final score. Take the arithmetic mean of the regional scores. Every region carries equal weight, so one slow region cannot be hidden by fast ones.

2.5 How much the success-rate cube matters

Cubing the success rate makes small reliability drops visible, but at the availability levels measured on BNB Chain the effect is small.

Success rate	Score multiplier
100.00%	1.000×
99.90%	1.003×
99.50%	1.015×
99.00%	1.031×
97.00%	1.096×
95.00%	1.166×

Availability on BNB Chain ranges from 99.95% to 100.00%, a multiplier of about 1.0015× or less. Alchemy posts 100.00% and still ranks second only by a 0.001 margin, because the formula weights latency directly. Success rate becomes a dominant term only well below 99%.

2.6 Reading the dashboard columns on BNB Chain

The P95 column is the mean of the three regional P95 values. Chainstack (97, 101, and 29 ms) averages 75.7 ms against 76 shown; Alchemy (75, 125, 178) averages 126.0 against 126; Quicknode (147, 107, 150) averages 134.7 against 135; dRPC (219, 248, 155) averages 207.3 against 207.

The score is not a simple function of that column. On BNB Chain, scores differ from mean P95 divided by availability cubed by between −4% and +15% (Chainstack: 0.087 against roughly 0.076), because the score averages each method’s own p95 while the percentile columns summarize a different aggregation. Rank order is identical under both measures. Treat the score as the ranking authority and the percentile columns as context.

2.7 Time window, limits, and independence

The dashboard is live, refreshes every 3 minutes, and offers 24-hour and 7-day views. Every figure here comes from the 7-day view captured on Oct 8, 2026. The benchmark does not measure end-user latency, throughput under load, response correctness, cost, or tiers above the entry paid plan. Chainstack chose the formula (p95, success rate cubed, equal regional weights) and operates the probes, so a different formula could reorder close pairs.

3. Providers and regions

Four providers with public BNB Chain RPC endpoints are monitored: Chainstack, Quicknode, Alchemy, and dRPC. Each is probed concurrently from DE, US, and SG over the same rolling 7-day window, on its first paid plan. To build on BNB Chain with Chainstack, see the Chainstack BNB Chain page.

4. Results

Table 1. BNB Chain, 7-day window, Oct 8, 2026. Chainstack leads on score and P95; Alchemy leads on P50; Quicknode and Alchemy lead Chainstack on P99.

Provider	Score	Availability	P50	P95	P99	DE	US	SG
Chainstack	0.087	99.98%	35ms	76ms	270ms	97ms	101ms	29ms
Alchemy	0.135	100.00%	32ms	126ms	222ms	75ms	125ms	178ms
Quicknode	0.136	99.99%	50ms	135ms	193ms	147ms	107ms	150ms
dRPC	0.200	99.95%	39ms	207ms	1,080ms	219ms	248ms	155ms

Chainstack’s lead over second place is 55% (0.087 against 0.135). Second and third place are effectively tied: a 0.001 difference is inside rounding.

5. Regional variance and outliers

The ranking and the tail tell different stories. Chainstack wins the score because its P95 (76 ms) is the lowest, but its P99 (270 ms) is higher than Quicknode’s (193 ms) and Alchemy’s (222 ms), and Alchemy’s P50 (32 ms) is the lowest in the set. A “best P50”, a “best P99”, and a “best score” answer different questions, and no provider leads all three. A workload sensitive to tail latency, such as liquidations or MEV-adjacent bots, should weigh the P99 column alongside the score.

Chainstack’s strongest cell is SG. Its 29 ms is the lowest single cell on BNB Chain. In DE, Alchemy (75 ms) beats Chainstack (97 ms), and in the US, Chainstack (101 ms) edges Quicknode (107 ms) by 6 ms.

Quicknode has the flattest profile but never leads a region. Its regional range is 107–150 ms (1.4×), against 29–101 ms for Chainstack (3.5×), 75–178 ms for Alchemy (2.4×), and 155–248 ms for dRPC (1.6×). A flat profile is only an advantage if its level is competitive, and Quicknode’s regional figures are all above the leader’s in their region.

Table 2. Lowest P95 by region on BNB Chain.

Region	Lowest P95
DE	Alchemy (75 ms)
US	Chainstack (101 ms)
SG	Chainstack (29 ms)

6. Changes since September

The September 2026 edition of this report covered the same four providers on BNB Chain. Table 3 compares its figures with this month’s.

Table 3. BNB Chain, September snapshot (Sep 28, 2026) to October snapshot (Oct 8, 2026). Lower is better.

Provider	Score	P95 (global)	P99 (global)	Worst region P95
Chainstack	0.091 → 0.087	93 → 76 ms	390 → 270 ms	US 148 → US 101 ms
Alchemy	0.124 → 0.135	122 → 126 ms	214 → 222 ms	US 143 → SG 178 ms
Quicknode	0.122 → 0.136	121 → 135 ms	268 → 193 ms	DE 147 → SG 150 ms
dRPC	0.419 → 0.200	464 → 207 ms	2,220 → 1,080 ms	US 689 → US 248 ms

dRPC’s September outlier did not persist. Its DE P95 fell from 559 ms to 219 ms and its US P95 from 689 ms to 248 ms, taking its score from 0.419 to 0.200. September showed dRPC with an unusual geography (weak in DE and US, strong in SG); October shows a more even profile.

Second and third place swapped. In September Quicknode (0.122) and Alchemy (0.124) were effectively tied; in October they are tied again at 0.136 and 0.135, in the opposite order. Alchemy’s SG P95 rose from 135 ms to 178 ms and Quicknode’s from 109 ms to 150 ms, which pushed both scores up.

Chainstack improved in every region except DE. Its US P95 fell from 148 ms to 101 ms and its SG from 41 ms to 29 ms, while DE rose from 89 ms to 97 ms. Its P99 fell from 390 ms to 270 ms but remains the weakest tail figure among the top three. Each snapshot is one 7-day window, so this shows two points in time, not a trend.

7. Interpretation notes and known confounders

Entry paid plan only. Higher tiers, including dedicated infrastructure placed near a weak region, could change a provider’s regional profile and are not tested.

Probes are synthetic and regional. Requests come from serverless functions in three cloud regions. A team whose users sit elsewhere should treat the nearest region as a proxy and verify with its own probes.

High throughput can widen the median-to-tail gap. On a high-volume chain, P99 says more about behavior under real load than on a quiet one. Chainstack’s P95 lead does not carry through to P99, so tail-sensitive workloads should not rely on the score alone.

Close scores are not rankings. Alchemy and Quicknode differ by 0.001. Read them as tied.

Method mix is summarized, not shown. Headline numbers average across methods with equal weights. Methods differ widely in cost, so a workload concentrated in one heavy method may rank providers differently. Per-method values are in Grafana.

Single window. Both the October figures and the September comparison rest on a single 7-day window each.

8. Conclusion

On BNB Chain, Chainstack posts the lowest score and the lowest P95, with its strongest region in SG. It does not lead every metric: Alchemy is faster at the median, and Quicknode and Alchemy are tighter at P99. dRPC recovered strongly from September, and second place is a statistical tie. Teams choosing an RPC provider for BNB Chain should decide which percentile matters for their workload (P50 for interactive responsiveness, P95 for the ranking used here, P99 for tail-sensitive systems) before picking a winner from Table 1, since no single provider leads all three this month.

9. Frequently asked questions

Which RPC provider has the lowest latency on BNB Chain?

Chainstack, on the 7-day window ending Oct 8, 2026. It scores 0.087 with a global P95 of 76 ms, against 126 ms for Alchemy, 135 ms for Quicknode, and 207 ms for dRPC.

Which provider has the best tail latency on BNB Chain?

Quicknode has the lowest P99 (193 ms), followed by Alchemy (222 ms), Chainstack (270 ms), and dRPC (1,080 ms). Workloads sensitive to tail latency should weigh P99 alongside the score.

How is the score calculated?

Score is response time divided by success rate cubed. It is computed per region from the mean of each method’s p95, then averaged across DE, US, and SG. Lower is better (Section 2.4).

Appendix A: measurement parameters

Parameter	Value
Chain	BNB Chain
Providers	Chainstack, Quicknode, Alchemy, dRPC
Plan monitored	First paid RPC plan per provider (Growth as a rule)
Regions	DE, US, SG
Probe infrastructure	Python service on Vercel serverless functions
Probe cadence	Every 1–3 minutes per region
Request target	Latest block; archive methods (⌛) query blocks 1–2 days old
Failure threshold	Responses over 55 seconds count as failures
Latency term	Mean of per-method p95, equal method weights, eth_subscribe excluded
Success-rate term	Sample-count weighted, eth_subscribe included
Score formula	ResponseTime / SuccessRate³ per region, then arithmetic mean across regions
Time window	Rolling 7 days; dashboard refreshes every 3 minutes
Snapshot date	Oct 8, 2026
Source	Chainstack Compare, public Grafana dashboard
Collection code	compare-dashboard-functions