# UWBench

UWBench is an open, vendor-neutral benchmark for AI agents that underwrite commercial credit. You can read it and run it yourself.

Checked on 11 November 2026.

## What UWBench is

UWBench is a test for AI agents that underwrite commercial loans. It is open. It is vendor-neutral. Anyone can read it. Anyone can run it.

It has four parts: a protocol, a case format, a trusted runner, and a scorecard that is mostly deterministic. The first public track uses synthetic US commercial term-loan cases. It has three lanes: raw documents, normalized data, and reasoning only.

## Who runs it

UWBench is an independent, member-based non-profit.

SecureLend helped initiate UWBench, has one board seat with no veto, and recuses from grading its own runs.

## What the results mean

UWBench is a public alpha. The version is v0.1.0-alpha.1. It was published on 20 August 2026.

The paper says the early measurements are engineering shakedowns. They are not certification results. They are not a controlled model ranking.

This page publishes no score for any vendor. It publishes no score for SecureLend.

## Run it yourself

Use UWBench when you want to judge an underwriting agent on a file, not on a demo. Do not treat a vendor chart as the test.

- Code on GitHub: https://github.com/uwbench/uwbench
- Release v0.1.0-alpha.1: https://github.com/uwbench/uwbench/releases/tag/v0.1.0-alpha.1
- Paper on Zenodo: https://doi.org/10.5281/zenodo.22035851

## Read more

SecureLend compares UWBench with other finance AI benchmarks: https://securelend.ai/blog/finance-ai-benchmarks-compared. That article does not report a blended SecureLend score.

SecureLend is a separate product. A deal file goes in. A cited credit memo comes out. A person makes the credit decision. It supports all commercial lending, including commercial mortgages and CRE, ABL, factoring, equipment finance, private credit, direct lending, and venture debt.