Every number here comes with the run that produced it

Four runs: what was measured, on how many samples, against what. Method under every card.

Start for free Docs

99.8%

TxStream delivered first against Yellowstone gRPC, over 103,073 shared transactions

0.032 ms

Hot Pump pair read on Quark, where Agave takes 33.7 s

330 ms

Average Beam landing time, against 457 ms on standard RPC

66%

Slots seen first on WebSocket against another provider, 257 observations by a third party

Streaming: the same transaction on two feeds, timed

Three streams from RPC Fast hosts into one collector, scored only on the transactions both feeds carried.

TxStream against Yellowstone gRPC

First on 99.8%

103,073 shared transactions, median lead shown. Same filter, one collector.

TxStream against Jito ShredStream

Four jobs less

TxStreamRaw shreds

Decoded transactions

Lookup tables resolved

Filters applied server-side

No shred receiver to run

30 µs apart at the median on 58,688 shared signatures. You get the finished transaction, not fragments.

Open the full report of this run →

17.08 mslead over Yellowstone gRPC at p95, on the same transaction.

Under 1 msadded with execution simulation on: 832 µs median for signatures only, 791 µs for the full payload.

0 msclient-side decoding. Transactions arrive decoded, lookup tables resolved and filters applied on our side.

Account reads: Quark against Agave, same call

Eight DeFi programs, the same call on both paths, Agave on the same gateway. Median shown.

Filtered getProgramAccounts

Agave, same gateway

Quark

Pump AMM pair

Agave, same gateway33.7 s

Quark

0.032 ms1,053,176× faster

Raydium liquidity v4 pair

Agave, same gateway5.48 s

Quark

0.026 ms210,871× faster

Raydium CLMM pair

Agave, same gateway9.56 s

Quark

0.245 ms39,025× faster

Orca Whirlpool pair

Agave, same gateway4.49 s

Quark

0.168 ms26,699× faster

Meteora DAMM v2 pair

Agave, same gateway6.91 s

Quark

0.273 ms25,314× faster

Meteora DLMM pair

Agave, same gateway4.05 s

Quark

0.270 ms14,994× faster

HumidiFi markets

Agave, same gateway53.96 ms

Quark

0.829 ms65.1× faster

Jupiter token ledgers

Agave, same gateway22.76 ms

Quark

10.46 ms2.18× faster

100 sampleson Quark and 3 on Agave per program, because a 33.7-second call is expensive to repeat. Both ends of the spread are published.

2.18×the smallest gain in the table, on a call Agave already answers in 23 ms. The rest of the spread is above.

100%of rows returned the same accounts as Agave on the same call.

›

Nothing changes in your client: same key, same methods. A filtered getProgramAccounts on a configured hot program goes to Quark, Agave stays underneath as the fallback. Quark is on the free Start tier too.

Landing: Beam against standard RPC

Time from submission to confirmation. Same wallets, same transactions, both paths landed them all.

Average time to land

127 ms faster

Same wallets, 20 runs per path, every transaction landed on both.

P90 time to land

476 ms faster

The slow tail is where a slot is missed. Same 20 runs.

Benchmark code, run it on your own key →

1 slotat most between submission and landing on Beam, against 4 on standard RPC in the same runs.

3 relaysinside Beam, each landed every transaction over 10 runs. The route is scored on the tail, so a relay with a bad P90 does not win.

Every plangets Beam, the free Start tier included. Same key, same sendTransaction.

WebSocket: who sees the slot first

The same slot subscription on two providers, measured by an independent third party from Frankfurt, not by us.

Slot subscription, RPC Fast against another provider

First on 66%

257 observations over three runs, measured by an independent third party, not by us.

The same test, run three times

Same direction every run

Frankfurt, one collector for both feeds. 84% of slots first or tied across the three runs.

Third-party results, as published →

6 to 11 msthe lead the other provider had on the slots where it was first.

Same keyas JSON-RPC and the streams. slotSubscribe on the WebSocket endpoint of any plan, free Start included.

Not oursthe run, the collector and the numbers come from the third party, on their hardware, not ours.

Rerun this on your own key

Free Start gives you 1.5M credits a month, Beam and the Quark read path, with no card and no call. The only benchmark that counts is the one against the provider you pay today.

Start for free Ask in Telegram