Location changes the result
These studies ran in Frankfurt. Your deployment region and network route can change feed arrival order.
Evidence, with context
Solana RPC benchmarks.
Measure the path. Understand the result.
Explore published Solana feed comparisons with the date, region and methodology kept beside the numbers. Use these studies as a starting point for testing your own workload.
Published studies
The result and its boundaries.
These are dated observations. They are not live measurements or a universal provider ranking.
2026-07-27 / Frankfurt
Shredstream vs Jito ShredStream
Raw shred delivery, first-detection latency
10,014transactions
Share of first detections
OrbitFlare95.93%
Jito ShredStream4.07%
A first detection is the first arrival among the feeds in this study. It is not a transaction landing rate.
Both feeds measured side by side on the same host in Frankfurt. Delays are per transaction, relative to whichever feed delivered it first. 10,014 valid transactions, zero backfill on either feed.
View the published sourceRead the numbers correctly
Same context. Useful comparisons.
A snapshot, not a promise
The March and July 2026 studies capture their own measurement windows. Historical Jito results do not imply current service availability.
Measure your whole pipeline
Delivery speed is one part of an application. Include decoding, data completeness, retries and cost in your evaluation.
The full methodology
Methodology version 1.0 · last updated 2026-07-27
Each measurement is taken by a probe running on a server inside one of our regions, so the figures reflect what a customer co-located with us would see rather than what our own racks see. Probes run against the same target set at the same time, on the same schedule, so rows in one run are directly comparable.
RPC request latency
Round-trip time for a fixed set of read methods (getLatestBlockhash, getAccountInfo on a warm account, getSlot) sent sequentially with a one-second gap, over a persistent HTTPS connection, for at least one hour per run. We report p50, p95 and p99 of the run and discard the first 30 seconds as warm-up.
Time to first shred
The gap between the leader’s first shred for a slot (timestamped from the Turbine broadcast received by an independent listener in the same region) and the moment each target’s feed delivers the first transaction from that slot to the probe. Measured per slot over the run and reported as percentiles. Slots where the leader skipped are excluded.
Shred delivery, side by side
Two raw shred feeds are consumed on the same host in the same region. For every valid transaction seen on both, we record which feed delivered it first and how far behind the other feed was. We report the share of transactions each feed delivered first (win rate) and the percentiles of the delay behind the first feed. A feed that is first on a transaction has a delay of zero for it. Transactions seen on only one feed, and any backfilled data, are excluded.
Transaction landing rate
A minimal self-transfer is signed against a fresh blockhash and sent through each target every five seconds. A transaction counts as landed if it is confirmed within the slot it was sent for or the following slot; later confirmations and expiries count as not landed. Reported as the share landed over the run, grouped by the region of the scheduled leader at send time.
Exclusions and versioning
Runs taken during a target’s announced maintenance window are dropped. Network-wide outages that affect every target are noted in the run and retained. Any change to the method bumps the methodology version; rows from different versions are never shown in the same table.
Sources
- Shredstream vs Jito ShredStream, Frankfurt, July 2026 (published chart)
- GeyserBench run c72e86ac (independent, verified), 1 March 2026
Background reading
- What is Shredstream: why the first shred is the earliest observation point.
- Network infrastructure for low-latency Solana trading: where the milliseconds go.
- Provider comparisons: interfaces, deployment and pricing side by side.
Your evaluation
Test what your application actually does.
Hold the setup constant
Use the same host, region and representative workload. Record the test window and client settings.
Measure more than a median
Track tail latency, missing events, reconnections, error rate and total spend.
Document the trade-offs
Choose a service based on the data your application needs and the results you can reproduce.
Questions about the data
Common questions.
What exactly is measured?
Three things, each from probes we run in our own regions against the live leader schedule: round-trip latency of common RPC methods, time from a leader broadcasting a block's first shred to our feed delivering it, and the share of test transactions that land in the slot they were sent for. Every row states the region, the target, the sample size and the date.
Why is the page marked as measurements in progress?
Because we publish numbers only when we have them. Until a measurement run has been completed, reviewed and added to the dataset, the page stays out of search indexes rather than show placeholder figures. The methodology is published first so the numbers can be checked against it when they arrive.
Can I reproduce the results?
Yes. The methodology section describes the probe design, the sample sizes and the exclusion rules. Run the same method from your own location and you should see the same ordering between providers, with absolute figures that differ by your distance to the leaders.
How often is the dataset updated?
Each published run carries its own date. We add runs when something changes: a new region, a client upgrade, or a competitor change worth re-measuring. The methodology version is bumped whenever the method itself changes, so rows from different versions are never compared silently.
Code. Build. Stream. Without Limits.
Bring your workload to the comparison.
Start with a test endpoint or discuss a streaming evaluation.
