GLM 5.2 is (nearly) as accurate as a human book-keeper at less than 1% of the cost
Summary
GLM 5.2 was benchmarked on VAT return preparation for a UK SME, claiming near-human accuracy at a fraction of the cost. The test processed 59 transactions in 68 minutes with a total token cost of 2.73 USD, yielding an almost-correct VAT return with a 7p discrepancy; mistakes included a serious share capital misclassification and some VAT category confusions. The article argues bookkeeping with AI is becoming a solvable problem for SMEs and highlights the need for scaffolding and human review for complex cases.