TECH5 Demonstrates a New Balance of Accuracy, Speed and Efficiency in NIST IREX 10

Latest TECH5 iris recognition algorithm records the fastest search time in NIST’s current two-eye developer table and one of the most compact templates measured, with competitive two-eye identification accuracy

August 13th, 2026, Geneva, Switzerland. TECH5, an innovator in the field of biometrics and digital identity, has demonstrated a distinctive balance of accuracy, speed and efficiency in the National Institute of Standards and Technology (NIST) Iris Exchange (IREX) 10 Identification Track. TECH5’s latest iris recognition submission, tech5_009, delivers competitive two-eye identification accuracy alongside the lowest search time currently reported in NIST’s two-eye developer table and a compact 6,237-byte template. The result is engineered for deployment at national scale, the demands of Digital Public Infrastructure (DPI), where accuracy alone does not determine whether a system is practical.

In NIST’s current two-eye evaluation, tech5_009 records a search time of just 0.088 ± 0.002 seconds against an enrolled population of 500,000 people, with an enrollment template size of 6,237 ± 99 bytes and a False Negative Identification Rate (FNIR) of 0.0067 ± 0.0005 at a False Positive Identification Rate (FPIR) of 0.01. Its 88-millisecond result is the lowest search time currently reported in NIST’s “Two-eye by Developer” table.

Looking beyond accuracy alone

Accuracy is fundamental to biometric identification, but it is not the only parameter that determines whether an algorithm is practical at national scale.

Large biometric systems must also address transaction latency, throughput, compute requirements, memory and storage footprint, infrastructure scalability, and ultimately total cost of ownership.

This is where the latest TECH5 result becomes particularly significant.

Compared with the five algorithms currently ranked highest by two-eye accuracy, NIST reports search times ranging from 4.6 to 12 seconds, compared with 0.088 seconds for TECH5. On the published benchmark figures, that makes TECH5 approximately 52× to 136× faster in the identification search.

The difference is similarly substantial for biometric template size. The five highest-accuracy entries use templates ranging from 18,280 to 130,879 bytes, while the TECH5 template is 6,237 bytes. TECH5 therefore uses approximately 66% to 95% less template storage than each of those five entries.

Even when compared specifically with the algorithm currently ranked first for two-eye accuracy—which, like the latest TECH5 submission, falls under NIST’s newer test platform—the published figures show 4.6 seconds versus 0.088 seconds of search time, and 19,349 bytes versus 6,237 bytes per template. That corresponds to approximately 52× shorter measured search time and 68% smaller templates for TECH5.

What milliseconds can mean at a border

Consider a busy port of entry using iris recognition to identify travelers against a large watchlist or identity repository.

In NIST’s 500,000-person identification benchmark, the matching stage using TECH5 completes in approximately 88 milliseconds. The corresponding published search times for the five most accurate algorithms range from 4.6 to 12 seconds.

Those figures should not be interpreted as complete passenger-processing times: image capture, networking, application logic, database access and operational workflows all contribute to end-to-end latency. But when biometric matching is performed repeatedly across multiple gates, lanes and simultaneous transactions, reducing the matcher component from seconds to milliseconds can materially reduce a potential source of transaction friction and make high-throughput architectures easier to design.

Smaller templates can change infrastructure economics

Template size becomes increasingly important as biometric databases move from hundreds of thousands to tens or hundreds of millions of identities.

As a simple illustration, extrapolating NIST’s reported template sizes to a hypothetical 100-million-person gallery, TECH5’s 6,237-byte template would represent approximately 624 GB of raw biometric template data. Using the template sizes reported for the five highest-accuracy entries, the same calculation produces approximately 1.83 TB to 13.09 TB.

These figures represent raw template payloads rather than actual production server requirements. Real systems require database structures, indexes, redundancy, backups and other overhead. Nevertheless, template size directly affects the amount of biometric data that must be stored, loaded, cached, replicated and transferred.

Combined with substantially faster search, a smaller template can potentially translate into fewer compute resources for a given transaction throughput, lower memory and storage requirements, simpler scaling, and lower infrastructure costs. The precise savings will depend on each deployment architecture and service-level requirements.

Optimizing the biometric system, not a single metric

The NIST results illustrate an important distinction between optimizing an algorithm for a single benchmark dimension and optimizing a biometric engine for deployment.

What the latest submission demonstrates is a deliberate engineering balance: competitive identification accuracy combined with an unusually low search latency and a compact template representation. The full comparative results are published by NIST. It reflects how TECH5 strives to optimize each technology component that goes into building national-scale DPI and, with it, the success of the programs that depend on it.

For high-volume applications such as border management, national identity, large-scale deduplication and inclusive civil registration — programs that must enroll and deduplicate entire populations while leaving citizens in control of their own identity data, and other population-scale identification systems, that balance can be as consequential as optimizing accuracy alone.

The results are publicly available on NIST’s website.

“With this NIST IREX 10 submission we improved the identification accuracy of our iris algorithm while recording the lowest search time in NIST’s two-eye developer table and a notably compact template. At national scale, that balance of speed and efficiency is as consequential as accuracy alone, and it is what makes population-scale identity systems practical to deploy and operate.” — Rahul Parthe, Co-Founder, Chairman and CTO, TECH5.

The algorithm is available to partners and customers worldwide as part of the T5-OmniMach product family.