Arc RPC Provider Benchmark: Latency & Uptime (September 2026)

By Chainstack Labs Research Team · Research Note No. 044 · September 2026 edition · Published Sep 21, 2026 · Last updated Sep 21, 2026 · Node Infrastructure

Data source: live Grafana compare dashboards, refreshed every 3 minutes, 7-day rolling window across DE, US, and SG.

Abstract

Arc is young enough as a network that public RPC provider comparisons barely exist yet, and the ones that do rarely account for region. This report ranks four RPC providers — Chainstack, Quicknode, Alchemy, and dRPC — on Arc RPC latency and uptime, using a single composite score that weights both response time and success rate, computed independently in three regions (Frankfurt, US, Singapore) and then averaged, so that a provider cannot post a strong global number by being fast in one region and absent in another. Chainstack ranks first with a score of 0.063, driven less by its raw p50 than by holding sub-80ms p95 latency consistently across all three regions; the next three providers each show at least one region where latency triples or worse, which the composite score punishes far more heavily than a simple global-average metric would. For a deeper look at how RPC method choice — not just provider or chain — drives latency, see our companion report, RPC Method Latency Across Chains.

Keywords: Arc RPC provider, Arc chain RPC latency, RPC benchmark, regional latency, availability, composite scoring, tail latency, provider comparison

Contents

1. Introduction
2. Methodology and scoring
3. Providers and regions
4. Results
5. Regional variance and outliers
6. Interpretation notes and known confounders
7. Conclusion
8. Appendix A: measurement parameters

1. Introduction

Most public RPC comparisons report one global latency number per provider, which hides the failure mode that actually hurts integrators building on Arc: a provider that is fast where it has infrastructure and slow or unreliable everywhere else. A trading bot running from Singapore does not care that a provider posts a strong number measured from a US test node, and a dashboard aggregating both without disclosing the split is not describing the same product to both users. This report ranks four RPC providers on Arc — Chainstack, Quicknode, Alchemy, and dRPC — using a score built to surface exactly that gap, and publishes the regional detail behind the ranking rather than only the final number, so a reader building for one region can override the global ranking with the row that actually applies to them.

The choice of provider matters more on a network like Arc than on a mature chain with a deep bench of RPC vendors: fewer providers currently run dedicated Arc infrastructure, so the spread between the best and worst option is wider than it would be on, say, Ethereum mainnet, where a decade of competition has narrowed the gap between vendors to a few milliseconds at the median.

2. Methodology and scoring

Each provider is evaluated on speed (response time) and reliability (success rate) independently in each monitored region, then combined as:

Score = ResponseTime / SuccessRate³, averaged across regions. Lower is better.

Cubing the success-rate term makes reliability the dominant factor once availability drops meaningfully below 100% — a provider that is fast but flaky is penalized far more than the raw percentage-point gap in its availability figure would suggest. For example, a provider at 99.88% availability does not read as “0.1 percentage points worse” under this formula; that gap already moves dRPC’s score enough to separate it from a provider with near-identical latency but higher availability. Per-method p95 latencies are combined with equal weights to produce the response-time term, and regional scores are averaged arithmetically rather than weighted by traffic share, so a slow region cannot be diluted by two fast ones. Scores are read directly from the underlying Grafana compare dashboards, which update every 3 minutes, rather than recomputed downstream from a static export — the numbers in this report are a snapshot of that live view, not a separately batched dataset.

3. Providers and regions

Four providers offering public Arc RPC endpoints were included: Chainstack, Quicknode, Alchemy, and dRPC. Each was measured concurrently from three regions — Frankfurt (DE), a US test node, and Singapore (SG) — against the same Arc mainnet RPC methods, over a rolling 7-day window refreshed every 3 minutes. No archive-node or dedicated-node tiers were included; all figures reflect each provider’s standard shared endpoint, which is also the tier most teams reach for first when evaluating an RPC provider for Arc before committing to a paid dedicated setup. Chainstack’s own Arc node deployment options include dedicated and elastic tiers that were out of scope for this round but are a natural next comparison once shared-tier rankings stabilize.

4. Results

Table 1. Composite score, availability, and global percentile latency by provider, 7-day window. Chainstack posts the lowest composite score (0.063) and the tightest global P95 (62ms); dRPC posts both the highest score and the only sub-99.9% availability in the set.

Table 2. P95 latency by region, milliseconds. Chainstack is the only provider with no region above 79ms; Quicknode and Alchemy each post the fastest single-region number in the table (22ms and 24ms in DE) but the widest single-region gap when Singapore is included.

5. Regional variance and outliers

The global P50/P95/P99 columns in Table 1 understate how uneven each provider’s regional footprint is. Quicknode’s 22ms DE p95 looks competitive with Chainstack’s until Table 2 shows the same provider at 177ms from Singapore — an eightfold regional spread that a single blended average would hide entirely. Alchemy shows the sharpest single-region cliff in the set: a 24ms P95 from Frankfurt against 382ms from Singapore, consistent with a footprint concentrated in the US and EU with comparatively thin routing to Southeast Asia. dRPC is the only provider whose composite score is also pulled down by availability (99.88% versus 99.97%+ for the other three), compounding its already-elevated latency across all three regions rather than one.

Chainstack is the only provider in the set with no region above 79ms P95, which is what produces its 0.063 score: not the lowest single-region number in the table, but the narrowest spread between best and worst region. This distinction — consistency versus peak speed — is the point of scoring on regional variance rather than a global average, and it is also the more actionable read for most teams: a provider that is occasionally the fastest option is less useful in production than one that is reliably good everywhere your users are.

6. Interpretation notes and known confounders

Shared-endpoint tier only. All figures reflect each provider’s standard shared RPC endpoint. Dedicated or archive-node tiers, where offered, were out of scope and would likely narrow the regional spread reported here, particularly for providers whose shared-tier footprint is concentrated in fewer regions — a dedicated node deployed directly in Singapore, for instance, would be expected to close most of the Alchemy and Quicknode SG gap reported in Table 2.

Arc is an early-stage network. Arc’s validator set, block times, and traffic profile are still maturing. Absolute latency figures for all four providers may shift as the network’s own infrastructure stabilizes, independent of any change in provider performance, which is why this report is dated and will be revisited rather than treated as a one-time result.

Method mix is not disclosed. The P95 figures combine multiple RPC methods with equal weights. As our RPC Method Latency Across Chains benchmark shows for other chains, method choice can move latency by an order of magnitude; a workload skewed toward one heavy method (e.g. log queries) may see different relative rankings than this blended figure suggests, and teams with a known method-heavy workload should weight that companion report’s findings alongside this one.

Single measurement window. Figures reflect one rolling 7-day window. The source dashboard also exposes a 24-hour view; day-to-day volatility, especially for dRPC’s wider percentile spread, was not characterized here and would benefit from a multi-week trend rather than a single snapshot. This report will be updated as that longer trend accumulates.

7. Conclusion

On Arc, ranking providers by regional consistency rather than a single global average changes the picture: Quicknode and Alchemy both post at least one region competitive with or better than Chainstack, but neither holds that performance across all three regions, and the composite score is built specifically to penalize that gap. For integrators serving users from a single region, the region-level table in Section 4 is more decision-relevant than the global ranking; for anyone serving traffic globally, Chainstack’s narrow best-to-worst-region spread is the more reliable signal. Teams choosing an RPC provider for Arc today should weigh which of those two profiles — peak regional speed or global consistency — matches where their own users actually are, and revisit this ranking as Arc’s own infrastructure and each provider’s regional footprint continue to mature.

Appendix A: measurement parameters

This is the September 2026 edition of a report Chainstack Labs plans to republish periodically as Arc’s network infrastructure and each provider’s regional footprint mature. It reflects a snapshot of the live Grafana compare dashboards described in Appendix A, refreshed every 3 minutes; figures may shift between the time of writing and the time of reading. See also: RPC Method Latency Across Chains, Arc node deployment options on Chainstack.