Benchmarks

Reproducible performance data. Every number on this page is generated by an open benchmark script available at github.com/quantum-elixir-tech/benchmarks — clone it and run against your own sandbox org to confirm.

Last updated: 2026-05-01. Benchmarks regenerate quarterly.

Latency

p50 / p95 / p99 against sandbox in Jakarta (production has lower jitter):

Identity

Endpoint	p50	p95	p99
`POST /api/identity/document/ktp/challenge`	18 ms	41 ms	78 ms
`POST /api/identity/document/ktp/capture`	612 ms	1180 ms	1890 ms
`POST /api/identity/face/flash-challenge`	14 ms	35 ms	64 ms
`POST /api/identity/face/liveness`	481 ms	920 ms	1410 ms
`POST /api/identity/face/match` (pool, 5 refs)	92 ms	174 ms	286 ms
`POST /api/identity/auth/step-up` (combined)	587 ms	1080 ms	1640 ms

AML

Endpoint	p50	p95	p99
`POST /api/screenings` (single subject)	124 ms	245 ms	384 ms
`POST /api/screenings` (batch of 100)	1284 ms	2110 ms	3104 ms
`POST /api/transactions/evaluate`	48 ms	92 ms	168 ms
`GET /api/alerts?status=open&limit=20`	31 ms	62 ms	124 ms

Anti-Fraud

Endpoint	p50	p95	p99
`POST /api/evaluate` (transaction lane)	41 ms	82 ms	154 ms
`POST /api/evaluate` (onboarding lane)	87 ms	168 ms	287 ms
`POST /api/device-session`	28 ms	58 ms	102 ms
`GET /api/sdk-config`	12 ms	24 ms	41 ms

Document Intelligence

Endpoint	p50	p95	p99
`POST /api/documents/upload` (multipart, 500 KB PDF)	184 ms	312 ms	481 ms
Async extract — KTP	2.1 s	3.4 s	5.2 s
Async extract — NPWP	1.8 s	2.9 s	4.4 s
Async extract — Invoice (single page)	2.4 s	3.8 s	5.7 s
Async extract — Contract (10 pages)	8.2 s	14.1 s	21.4 s

Bank Statement

Operation	p50	p95	p99
Upload (1 MB BCA statement)	142 ms	248 ms	384 ms
Async parse — BCA native	1.2 s	2.1 s	3.4 s
Async parse — Mandiri native	1.1 s	1.9 s	3.1 s
Async parse — AI fallback (unsupported bank)	4.8 s	7.2 s	11.4 s
`GET /api/consolidations/{id}/consolidated` (5 statements)	87 ms	184 ms	312 ms

Orchestration / AI Automation

Operation	p50	p95	p99
Workflow execute (trigger to first step)	124 ms	234 ms	384 ms
Per `service_call` step overhead	18 ms	41 ms	78 ms
LLM step (Quantum AI default, ~500 tokens)	1.2 s	2.4 s	3.8 s
Cron schedule jitter	12 s	47 s	58 s

Throughput

Per-org throughput on the standard tier. Higher tiers scale linearly.

Endpoint class	Standard	Premium	Enterprise
Reads (`GET`)	10,000 RPM	30,000 RPM	100,000 RPM
Transaction evaluate (`POST /api/transactions/evaluate` + `POST /api/evaluate`)	1,000 RPM	5,000 RPM	20,000 RPM
Screenings	60 RPM	600 RPM	5,000 RPM
Document/statement uploads	60 / 10min	600 / 10min	unmetered

Accuracy

Cross-validated against held-out test sets curated with Indonesian financial-services partners.

Bank Statement parsing

Parser	Field accuracy (transactions)	Balance reconciliation	Bank detection accuracy
BCA native	99.6%	99.9%	99.8%
Mandiri native	99.4%	99.8%	99.7%
BRI native	99.2%	99.5%	99.6%
BNI native	99.3%	99.7%	99.6%
Permata native	98.8%	99.2%	99.3%
Danamon native	98.5%	99.0%	99.1%
BTN native	98.4%	98.7%	98.9%
OCBC native	98.6%	99.1%	99.0%
AI fallback	95.8%	97.3%	n/a

KTP capture (Identity Platform)

Check	True positive	False positive
Anti-spoof live vs print	99.4%	0.7%
Anti-spoof live vs digital screen	98.9%	1.2%
Anti-spoof live vs 3D mask (premium)	97.2%	1.8%
Tamper detection (modified NIK)	99.8%	0.2%
Face match (same person, KTP vs selfie)	98.6%	0.3%

AML screening

Metric	Value
True-positive rate at default thresholds	98.7%
False-positive rate at default thresholds	0.4%
OFAC SDN coverage (vs. official source)	100% (hourly refresh)
Adverse-media corpus size	12M+ Indonesian-language articles

Methodology

All latency numbers:

Measured from same-region client (id-jkt-1 → id-jkt-1).
10,000-request samples per endpoint, distributed over a one-hour window.
Excludes connection establishment (assumes warm TLS).
Includes auth, validation, business logic, and DB round-trips.

Accuracy numbers:

Curated test sets (~5,000 examples per category) with human-labelled ground truth.
80/20 train/test split with the model never trained on the test partition.
Re-validated quarterly against new examples.

Reproduce yourself:

git clone https://github.com/quantum-elixir-tech/benchmarks
cd benchmarks
export QE_API_KEY=qe_sk_sandbox_...
npm install
npm run bench:identity   # or aml, anti-fraud, etc.

Production numbers will look slightly better

Sandbox is shared infrastructure. Production runs on dedicated capacity with no noisy neighbours. Expect ~20% better p95 in production than the numbers above. Premium tier guarantees stricter SLAs.

Status Overview