← Retour au blog
tech 10 July 2026

GLM 5.2: Accuracy Comparable to a Human Bookkeeper

GLM 5.2 redefines VAT return preparation with near-human accuracy at a fraction of the traditional cost.

Article inspired by the original source
GLM 5.2 is nearly as accurate as a human book keeper ↗ toot-books.pages.dev

Introduction

In the business world, accuracy is paramount, especially when it comes to accounting and tax compliance. With the rise of artificial intelligence, many businesses are turning to automated solutions to streamline these processes. GLM 5.2, an open weights AI model, has recently demonstrated its ability to match the accuracy of a human bookkeeper in preparing quarterly VAT returns for small UK businesses.

The Challenge of VAT Returns

Preparing a VAT return is a crucial task for any VAT-registered business in the UK. Small and medium-sized enterprises (SMEs) often rely on external accounting firms for this task. Typical fees for these services range from 750 to 2,100 GBP per quarter, or between 1,000 and 2,800 USD. Furthermore, strict deadlines require submissions within five weeks of the quarter's end to avoid substantial penalties.

GLM 5.2: An Economical and Accurate Solution

In a recent test, GLM 5.2 demonstrated it could prepare a nearly perfect VAT return for a UK SME, processing 59 transactions in 68 minutes at a raw token cost of only 2.73 USD. The model used a command-line tool (CLI) to input each transaction into the accounting software. The final output was evaluated for accuracy against six criteria per transaction. The model produced an essentially correct VAT return, with the net position off by only 7 pence (~10 US cents) relative to the ground truth.

Benchmark Methodology

To evaluate GLM 5.2, the model processed transaction data and corresponding receipts from Vineyard Finance's first quarter of 2026, initially prepared by humans. The human-prepared books went beyond what was asked of the model, involving the search for relevant invoices and the handling of specific circumstances not deducible from bank statements alone.

GLM 5.2 operated on an isolated Google Cloud Platform (GCP) instance, with access to the internet and cloud-based accounting software, as well as a pre-authenticated CLI tool. The model ran on a minimalistic setup that exposed only two tools: a bash tool and a session termination plus final reporting tool.

Advantages and Limitations of GLM 5.2

One of the primary advantages of GLM 5.2 is its extremely low operating cost. In comparison, the fees for a human accountant are significantly higher. However, while GLM 5.2 offers impressive accuracy, it does not entirely replace the need for human intervention, particularly in contexts where complex judgment is required.

Conclusion

GLM 5.2 represents a significant advancement in automating accounting tasks, providing SMEs with a precise and cost-effective solution for VAT return preparation. By reducing costs and increasing efficiency, this model could revolutionize how businesses handle their tax obligations.

Let's discuss your project in 15 minutes.

GLM 5.2 AI accounting VAT return automation cost-effective AI business automation
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call