--- title: GLM 5.2 is (nearly) as accurate as a human book-keeper... description: GLM 5.2, an open-weights AI model, was benchmarked on preparing a real quarterly VAT return for a UK SME. Running as an autonomous agent with CLI access to... canonical: https://daily.dev/posts/glm-5-2-is-nearly-as-accurate-as-a-human-book-keeper-at-less-than-1-of-the-cost-naggeyjs3 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: GLM 5.2 is (nearly) as accurate as a human book-keeper at less than 1% of the cost | daily.dev og:description: GLM 5.2, an open-weights AI model, was benchmarked on preparing a real quarterly VAT return for a UK SME. Running as an autonomous agent with CLI access to... og:url: https://daily.dev/posts/glm-5-2-is-nearly-as-accurate-as-a-human-book-keeper-at-less-than-1-of-the-cost-naggeyjs3 og:image: https://api.daily.dev/og/posts/naGgEYjs3.png og:image:alt: GLM 5.2 is (nearly) as accurate as a human book-keeper at less than 1% of the cost og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # GLM 5.2 is (nearly) as accurate as a human book-keeper at less than 1% of the cost **[Hacker News](https://daily.dev/sources/hn)** · 9 min read · 1 upvotes · 0 comments ## Summary GLM 5.2, an open-weights AI model, was benchmarked on preparing a real quarterly VAT return for a UK SME. Running as an autonomous agent with CLI access to accounting software, it processed 59 transactions in 68 minutes at a raw token cost of $2.73 — compared to the £750–2,100/quarter typically charged by human accountants. Out of 354 scored checks, the model failed 20. The most significant error was misclassifying founder share capital (£10,000) to the wrong account, which carries legal implications. It also confused zero-rated and VAT-exempt categories in 14 transactions. Despite these errors, the final VAT return figure (Box 5) was off by only 7 pence from the human-prepared ground truth. The post concludes that bookkeeping is becoming a solved problem and promotes an open beta product built on these capabilities. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How accurate was GLM 5.2 at preparing a real UK VAT return compared to a human bookkeeper? GLM 5.2 produced a VAT return whose net position (Box 5) was off by only 7 pence relative to the human-prepared ground truth, processing 59 real transactions from a UK SME's quarter of books. Out of 354 scored checks across 6 criteria per transaction, it failed only 20, with just one serious error involving misclassified founder share capital. _Developers evaluating AI agents for financial workflows can follow real-world accuracy benchmarks like this on daily.dev._ ### How much did it cost to run GLM 5.2 as an accounting agent for a quarter of bookkeeping? Processing 59 transactions across a full UK VAT quarter cost $2.73 in raw token spend and took 68 minutes of wall time, using the Fireworks AI serverless tier as the model provider. This compares to a typical human accounting firm fee of 750-2,100 GBP (roughly $1,000-2,800) per quarter for the same VAT return preparation task. _Track cost-versus-accuracy tradeoffs for AI agents replacing manual workflows on daily.dev._ ### What mistakes did GLM 5.2 make when doing automated bookkeeping and VAT categorization? The most serious error was booking a 10,000 GBP founder share capital payment to a generic capital account instead of the legally distinct unpaid shares account, which has audit and end-of-year filing implications. Fourteen other errors stemmed from confusing zero-rated VAT with tax-exempt VAT, and three involved mishandling a Wise multi-currency split transaction during VAT allocation. _Anyone building or auditing AI bookkeeping agents can compare failure patterns like these on daily.dev._ ## Similar posts on daily.dev - [Open-source AI is just “4 months behind” closed frontier models — and 10x cheaper](https://daily.dev/posts/open-source-ai-is-just-4-months-behind-closed-frontier-models-and-10x-cheaper-ogsojrepz) · The New Stack · 0 upvotes · 0 comments - [Comparing token economics across 42 AI models](https://daily.dev/posts/comparing-token-economics-across-42-ai-models-htlm2dnb1) · Neon · 2 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#ai-agents](https://daily.dev/tags/ai-agents), [#fintech](https://daily.dev/tags/fintech), [#glm](https://daily.dev/tags/glm) [View this post on daily.dev](https://daily.dev/posts/glm-5-2-is-nearly-as-accurate-as-a-human-book-keeper-at-less-than-1-of-the-cost-naggeyjs3) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"GLM 5.2 is (nearly) as accurate as a human book-keeper at less than 1% of the cost","url":"https://daily.dev/posts/glm-5-2-is-nearly-as-accurate-as-a-human-book-keeper-at-less-than-1-of-the-cost-naggeyjs3","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/glm-5-2-is-nearly-as-accurate-as-a-human-book-keeper-at-less-than-1-of-the-cost-naggeyjs3"},"datePublished":"2026-07-09T19:39:39.517Z","dateModified":"2026-09-14T08:00:54.060Z","description":"GLM 5.2, an open-weights AI model, was benchmarked on preparing a real quarterly VAT return for a UK SME. Running as an autonomous agent with CLI access to...","image":"https://media.daily.dev/image/upload/s--foaA6JGU--/f_auto/v1722860399/public/Placeholder%2004","thumbnailUrl":"https://media.daily.dev/image/upload/s--foaA6JGU--/f_auto/v1722860399/public/Placeholder%2004","isAccessibleForFree":true,"articleSection":"Hacker News","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hacker News","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/hn","url":"https://daily.dev/sources/hn"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/glm-5-2-is-nearly-as-accurate-as-a-human-book-keeper-at-less-than-1-of-the-cost-naggeyjs3","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,ai-agents,fintech,glm","timeRequired":"PT9M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hacker News","item":"https://daily.dev/sources/hn"},{"@type":"ListItem","position":3,"name":"GLM 5.2 is (nearly) as accurate as a human book-keeper at less than 1% of the cost"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/glm-5-2-is-nearly-as-accurate-as-a-human-book-keeper-at-less-than-1-of-the-cost-naggeyjs3#faq","mainEntity":[{"@type":"Question","name":"How accurate was GLM 5.2 at preparing a real UK VAT return compared to a human bookkeeper?","acceptedAnswer":{"@type":"Answer","text":"GLM 5.2 produced a VAT return whose net position (Box 5) was off by only 7 pence relative to the human-prepared ground truth, processing 59 real transactions from a UK SME's quarter of books. Out of 354 scored checks across 6 criteria per transaction, it failed only 20, with just one serious error involving misclassified founder share capital. Developers evaluating AI agents for financial workflows can follow real-world accuracy benchmarks like this on daily.dev."}},{"@type":"Question","name":"How much did it cost to run GLM 5.2 as an accounting agent for a quarter of bookkeeping?","acceptedAnswer":{"@type":"Answer","text":"Processing 59 transactions across a full UK VAT quarter cost $2.73 in raw token spend and took 68 minutes of wall time, using the Fireworks AI serverless tier as the model provider. This compares to a typical human accounting firm fee of 750-2,100 GBP (roughly $1,000-2,800) per quarter for the same VAT return preparation task. Track cost-versus-accuracy tradeoffs for AI agents replacing manual workflows on daily.dev."}},{"@type":"Question","name":"What mistakes did GLM 5.2 make when doing automated bookkeeping and VAT categorization?","acceptedAnswer":{"@type":"Answer","text":"The most serious error was booking a 10,000 GBP founder share capital payment to a generic capital account instead of the legally distinct unpaid shares account, which has audit and end-of-year filing implications. Fourteen other errors stemmed from confusing zero-rated VAT with tax-exempt VAT, and three involved mishandling a Wise multi-currency split transaction during VAT allocation. Anyone building or auditing AI bookkeeping agents can compare failure patterns like these on daily.dev."}}]} ```