RPC pricing, normalized: cost per 1M requests

Compute units, API credits, responses, requests — every provider invents its own billing unit, and the multipliers obscure the only number that matters for a budget: what does a million calls cost? Here’s the market normalized to that number, and the honest math on where flat-rate pricing beats metered — and where it doesn’t.

The market at entry-plan pricing

Effective cost of 1M standard eth_call-class requests on each provider’s entry paid plan, using each provider’s own published method weights (checked August 2026 — see methodology below):

The break-even column is the entire decision, compressed: if you make more requests per month than that number, flat-rate is cheaper than that provider — at any volume above it, dramatically so. Against the most expensive metered plans the crossover is under 3M requests/month, which a modest indexer does in a day.

What flat-rate costs per million, by volume

A metered provider’s cost per million is fixed. A flat plan’s falls with every request you make. SwiftNodes Starter ($49/mo, 50 req/s, no volume cap):

Higher tiers raise the rate ceiling, not the price curve’s shape: Growth ($89, 150 req/s) bottoms out around $0.23 per million; Pro ($249, 500 req/s) around $0.19. And because nothing is metered, a trace call costs the same as an eth_call— the method-weight multipliers that make compute-unit budgeting painful simply don’t exist here.

Where metered pricing honestly wins

• Low volume. Under a few million requests a month, most metered providers’ free tiers cover you at $0 — and our own free tier (250K/mo) is the honest choice before paying anyone.
• Feature needs beyond RPC. Alchemy’s NFT/token APIs, QuickNode’s streams and webhooks, Infura’s MetaMask integration — if those are why you’re there, price per call isn’t the deciding number.
• Extreme burst over low average. Flat plans meter rate (req/s), not volume. If you need 2,000 req/s for one hour a month and little else, a metered plan may fit better than a high-rate flat tier.

Methodology

Numbers come from each provider’s public pricing page as of August 2026, cross-checked against Dwellir’s independent normalization (a competitor’s comparison — we cite it because its per-call math checks out). Metered costs assume standard eth_call-class weights; trace/debug and archive-heavy workloads cost multiples more on compute-unit systems (see our deep dives on compute units and API credits). SwiftNodes’ effective-cost math divides the flat monthly price by your actual volume; the practical ceiling shown is the plan’s req/s limit sustained for a 30-day month. Prices change — if you spot a stale number, tell usand we’ll fix it.