Providers
The best Ethereum RPC providers, measured
Every other page answering this question is published by one of the providers being ranked. This one is not. 9 providers, 286,086 calls in the last 24 hours, from four regions — and the method is published so the ranking can be checked rather than trusted.
Tenderly leads
Tenderly scored 99.4 over 38,592 calls in the last 24 hours — 33 ms median response, 99.98% answered, 100.00% of answers correct. PublicNode is next at 97.7.
That is the ranking's own output, not an editorial pick. It changes when the measurements do.
All 9 ranked
Last 24 hours, every region combined. Lower latency is better; everything else is higher.
| # | Provider | Score | Median | Answered | Calls |
|---|---|---|---|---|---|
| 1 | Tenderly | 99.4 | 33 ms | 99.98% | 38,592 |
| 2 | PublicNodeno key needed | 97.7 | 42 ms | 99.99% | 38,960 |
| 3 | dRPC | 91.2 | 70 ms | 99.94% | 37,992 |
| 4 | NodeReal | 86.7 | 100 ms | 100.00% | 37,995 |
| 5 | Blast API | 83.9 | 123 ms | 99.97% | 38,541 |
| 6 | Merkle | 78.0 | 193 ms | 100.00% | 8,866 |
| 7 | MEV Blockerno key needed | 76.9 | 213 ms | 99.94% | 38,551 |
| 8 | 1RPCno key needed | 57.3 | 6590 ms | 72.66% | 7,948 |
| 9 | Cloudflareno key needed | 53.6 | 86 ms | 23.16% | 38,641 |
How to read this
- The score is not a review
- It is availability at 40%, latency at 30%, correctness at 20% and chain-head freshness at 10%, computed against the cohort measured in the same window. A provider measured on fewer calls is marked as such rather than flattered by a small sample.
- Latency depends on where you are
- These figures combine four regions. If your workload runs in one place, the nearest measured endpoints will tell you more than a global median will.
- We measure endpoints, not plans
- A commercial provider's paid tier may behave differently from the endpoint measured here. Nothing on this page is a claim about pricing, support or terms — only about what an endpoint did when it was called.
- Nobody paid to be here
- No provider pays for placement, none reviews its score before publication, and any of them can ask for a correction — which changes a measurement only if the measurement was wrong.