Icon Solutions MongoDB benchmark results landed on Tuesday, September 8, 2026. The two companies said the Icon Payments Framework (IPF), running on MongoDB Atlas, processed 6,000 SEPA Instant-style payments per second end to end. Mean latency was 0.51 seconds, well inside the roughly 10-second window instant schemes typically allow. Banks including Citi, NatWest and BNP Paribas already use IPF, so this is not a benchmark of a stack nobody has deployed.

What the Icon Solutions MongoDB Benchmark Measured

Each test payment ran through 12 processing steps, including duplicate checks, scheme-rule validation, fraud and sanctions screening, FX retrieval and booking. The 6,000 payments per second translated to about 430,000 database operations per second across two Atlas clusters. According to the joint release, persistence latency stayed below 30 milliseconds at peak.

The failure tests are the more interesting numbers. IPF recovered from an ungraceful application-node crash in under 90 seconds and from a planned node shutdown in about 60 seconds. Separately, an Atlas primary failover completed in about 5 seconds on average. The Icon Solutions MongoDB setup recorded zero data loss across every scenario tested. In addition, upgrading Atlas from 8.0 to 8.3 cut persistence latency by about half, from roughly 40 to 20 milliseconds at 4,000 payments per second.

Icon strategy director Toine van Beusekom framed the results as proof banks can handle always-on volumes without self-building. MongoDB’s Boris Bialek pointed to fast node recovery and headroom at peak.

The Icon Solutions MongoDB Caveat Comes First

Say the obvious thing first. This is the Icon Solutions MongoDB pairing benchmarking its own stack and publishing the results itself. There is no independent lab, no third-party auditor and no rival vendor running the same test on the same hardware. Treat every number as what the vendors say the product can do, not as verified fact.

What the Icon Solutions MongoDB Numbers Leave Out

The fine print adds more caveats. First, the zero-data-loss application failure tests ran at 3,500 payments per second, not at the 6,000 peak. Second, the bank’s surrounding systems, including fraud, sanctions and accounting, were simulators. Real downstream calls would add latency the headline figure does not capture.

Third, the release reports mean latency, not tail latency such as the 99th percentile. Tail latency is what decides whether individual payments breach scheme timeouts. Fourth, the setup ran across three availability zones in a single cloud region, so a full regional outage was not tested. Finally, it used 11 large Atlas shards, and the full dataset is available only on request.

Fintechbits analysis

Why Icon Solutions MongoDB Results Still Matter

With those caveats up front, the numbers are still worth attention. Instant schemes live or die on two questions. Can you clear payments fast enough under load, and what happens to customer money when a node dies mid-transaction?

A sub-second mean with zero data loss under both an application crash and a database failover is a relevant data point for banks shopping for instant-payments infrastructure. That matters now that EU rules require euro-area payment providers to offer instant transfers, and as U.S. banks weigh FedNow. The 8.0-to-8.3 improvement is also a useful nudge for banks running older MongoDB versions, as instant rails reshape the future of payments.

What to Watch After the Icon Solutions MongoDB Benchmark

The real test of the Icon Solutions MongoDB claims will be a live production incident at a bank customer, not another controlled benchmark. Watch whether Citi, NatWest or BNP Paribas discuss instant-payments uptime publicly over the next year. Also watch whether Icon releases tail-latency data or invites an independent test. Operational resilience rules make that kind of evidence increasingly important, a theme central to regtech.

Fintechbits covers payments infrastructure, instant payments and banking technology. Nothing here constitutes financial or investment advice. All analysis represents the editorial views of Fintechbits.