Press Release · Xailient
Xailient Ranks #1 for Face Recognition Accuracy in Real-world On-Device AI
Distributed by Endless Tech · Originally issued via PR Newswire
Independent NIST testing shows Xailient delivers leading Border-Kiosk accuracy among small-and-fast algorithms.
LAS VEGAS, Nov. 17, 2026 — Xailient, a developer of edge AI face recognition technology, today announced that its latest face recognition model achieved the highest accuracy in the U.S. National Institute of Standards and Technology (NIST) Face Recognition Technology Evaluation (FRTE) Border-Kiosk test among 106 developers whose template generation time is 500 milliseconds (ms) or less.
In contrast to server-based or hub-based Face Recognition Technology (FRT) which have no fixed limit, On-Device FRT must fit within the constraints of 1 GigaByte (GB) of RAM and 500ms response time in the NIST benchmark systems.
NIST tested the same model, xailient_001, that Xailient deploys on edge hardware with approximately 1 TOPS of NPU compute. Running recognition on-device reduces dependence on cloud connectivity and allows face matching to occur at the point of capture.
The result highlights a unique strength of Xailient's technology: high face recognition accuracy under real-world conditions without large compute budgets or cloud-scale infrastructure.
Why Border-Kiosk matters
NIST's Border-Kiosk test is the most relevant to real-world deployments because it is the only static-image NIST 1:1 test involving non-cooperative subjects. Subjects interact with a touchscreen positioned below the camera, creating difficult viewing angles including downward facial pitch that can exceed 45 degrees. Kiosk images are compared with border-crossing images, such as those taken when walking through a gate or standing before a border agent - thereby matching two "unposed" images.
#1 of 106 - Leading accuracy without leading compute
At a false match rate of 1 in 100,000, xailient_001 correctly matched 95.4% of genuine Border-Kiosk image pairs while operating at a low false-match rate - the best result in the category.
NIST testing sets a benchmark false match rate, and records the non-match accuracy. At an even stricter false match rate of 1 in 1,000,000, Xailient recorded an equivalent to 94.0% of genuine pairs correctly matched. On the easier Border-Border test, Xailient achieved a 0.55% false non-match rate at a false match rate of 1 in 1,000,000 — approximately 99.45% genuine-match accuracy at an exceptionally strict operating point. In production, Xailient employs multiple methods to achieve a much lower false match rate.
Xailient generated face templates in 497 milliseconds in NIST testing. In Xailient's comparison of six major competitors, the model was 1.6× faster and produced 12% fewer missed matches than close competitors. Xailient used 901 MB of memory and 280 MB of configuration data, the lowest figures in the comparison set. The nearest competitor required 1.8x more memory and 1.6x more configuration data. The largest competitor systems required 4x more compute and up to 6.7x more memory and 5.7x more configuration data than Xailient.
In practical terms, lower compute requirements translate to cost savings when selecting AI capable chips, such as Neural Processing Units (NPUs). Lower memory reduces the RAM and flash storage required on each device. Device costs have been rising, driven in part by RAM price increases of 400% since 2024 and Flash memory price increases of 250%. More efficient software translates to cost savings.
Efficiency critical to on-Device
On-Device AI must operate within constraints of power and cost, and xailient_001 runs directly on cameras using NPUs rated at approximately 1 TOPS, enabling sophisticated face recognition at the edge without requiring transmission of sensitive data or expensive GPU infrastructure.
- "Face recognition performance in the real world is not just about achieving the best score with unlimited compute," said Lars Oleson, CEO of Xailient. "It is about delivering high accuracy on the hardware customers actually deploy. These results prove that Xailient can deliver real world accuracy while operating inside a dramatically smaller compute envelope."
- "Edge deployment isn't just about cost efficiency, it is also essential for privacy. By matching on the camera, edge deployment avoids collecting bystander data - only recognizing opt-in and officially registered subjects," said Mallika Patel, VP of Product.
Real-world performance can exceed single-shot benchmarks
NIST FRTE 1:1 evaluates one image pair at a time. Xailient's production system evaluates multiple frames as a subject approaches or interacts with the device, so recognition is based on repeated observations rather than a single frame. Xailient combines those repeated observations with image-quality gating, matching thresholds and application-level verification logic to improve recognition reliability and reduce false positives.
The NIST lab result arrives on the back of recent independent ISO live-subject testing of Xailient systems. In that testing 250 diverse, moving subjects were matched against a gallery of 10,000 registered candidates in real-time. Xailient achieved 99.5% successful recognition with zero false matches.
From benchmark to deployment
For customers in gaming, access control, retail and other edge environments, the combination of accuracy, speed and compact model size means face recognition can be deployed directly at the point of capture with lower hardware requirements, reduced latency, and greater control over data processing.
Notes to editors
NIST FRTE is an independent, ongoing evaluation of face recognition algorithms conducted by the U.S. National Institute of Standards and Technology. Xailient's November 2026 submission is identified as xailient_001.
Results shown from NIST do not constitute an endorsement of any particular system, product, service, or company by NIST.
NIST Face Recognition Technology Evaluation: https://www.nist.gov/programs-projects/face-recognition-vendor-test-frvt-ongoing
SOURCE Xailient
About Xailient
Xailient develops edge AI face recognition technology designed for high-accuracy, low-latency deployment on resource-constrained hardware. Its technology enables face recognition directly on cameras and edge devices for applications including gaming, access control, security and retail.