Performance & Benchmarks

Open benchmarks.
All runs linked.

gRPC streaming latency. Transaction landing times. Full distribution — P50, P95, P99. Every test has a date, a node, a run count, and a public link.

Start for Free Read the Docs Aperture · Beam · gRPC benchmarks

Aperture · gRPC Streaming

First-look in 99.97%
of slots vs Yellowstone.

Aperture is a shred-native gRPC feed. We ran 20,000 transactions head-to-head against Yellowstone and measured which feed saw each transaction first — and by how much.

Aperture vs Yellowstone · 20,000 tx

Aperture — winner

First-look slots

99.97%

19,994 / 20,000 transactions

Avg lead when first

10.7ms

P95 lead

35.5ms

P99 lead

50.7ms

Yellowstone — standard gRPC

First-look slots

0.03%

6 / 20,000 transactions

P50 deficit

10.77ms

P95 deficit

35.56ms

P99 deficit

50.72ms

First-look detection rate

% of 20,000 slots where each feed was first

Aperture advantage over Yellowstone — latency distribution

How many ms ahead Aperture was when it won first-look · P50 / P95 / P99

Aperture vs Shredstream — context

Overall latency difference across 20,000 slots: Shredstream is only 0.1ms faster on average, 0.4ms at P95, 2.1ms at the tail. The gap is this small because Aperture runs server-side filters and reduced bandwidth — it does more work on the same latency budget.

Shredstream · raw firehose

First-look slots 95.59%

Overall avg advantage 0.1ms

Overall P95 advantage 0.4ms

Overall P99 advantage 2.1ms

Server-side filters No

Bandwidth Full stream

Aperture · filtered feed

First-look slots 4.41%

Overall avg deficit 0.1ms

Overall P95 deficit 0.4ms

Overall P99 deficit 2.1ms

Server-side filters Yes

Bandwidth Reduced

Full runs: Aperture vs Yellowstone ↗ Aperture vs Shredstream ↗

Beam · Transaction Landing

Routes to the fastest
provider on every run.

Beam scores providers in real time and routes each transaction to the top performer. P90 is weighted 4× in the scoring — tail latency is where landing fails under load. Available on all paid plans (extra fees via partners' network).

Provider scoring · P90-weighted 4× · 10 runs each

beam-falcon avg 229ms · P90 444ms · ack 54ms

100.00%

beam-astralane avg 231ms · P90 444ms · ack 52ms

99.78%

beam-bloxroute avg 233ms · P90 444ms · ack 53ms

99.66%

All three providers: 100% success ratio · avg landed slots 0.7 · P90 landed slots 1.0

Beam vs standard RPC · 20 runs

Provider	Avg landed	P90 landed	Performance score	Avg ack
beam-astralane	330ms	589ms	97.42%	53ms
rpcfast-rpc (standard)	457ms	1,065ms	82.20%	28ms

Beam: −127ms avg · P90 gap −476ms. Standard RPC acks faster (no routing overhead) but landing latency is 38% higher.

Beam vs standard RPC — landing latency

Avg and P90 landing time · 20 runs

Test Conditions

Public methodology.
No exceptions.

Every number includes the test date, node spec, run count, and scoring formula. Runs are linked — check them yourself.

Test Environment

Date May 19, 2026
Location Frankfurt, Germany
Node Latitude.sh — f4.metal.small
gRPC runs 20,000 tx per feed
Beam scoring 10 runs per provider
Beam vs RPC 20 runs per provider

Beam Scoring Formula

landed_ms 0.2 × avg + 0.8 × P90 (P90 weighted 4×)
landed_slots 0.2 × avg + 0.8 × P90
landed_idx 0.5 × avg + 0.5 × P90
same_slot higher is better (0–100)
success_ratio higher is better (0–100)
Final score unweighted average of all five

Run: Aperture vs Yellowstone ↗
Run: Aperture vs Shredstream ↗

Open benchmarks.All runs linked.

First-look in 99.97%of slots vs Yellowstone.