Claude Opus 4.8 vs GPT-5.3-Codex for bookkeeping

The verdict

GPT-5.3-Codex wins, 9283

Same 12 bookkeeping tasks, marked blind by three rival labs. GPT-5.3-Codex took 3 tasks, Claude Opus 4.8 took 1, 8 tied. Tested 18 Aug 2026.

Where they differed most

The task — Find the £36

A bank reconciliation will not balance. The bank statement closing balance is £8,412.55. The cash book closing balance is £8,376.55. Outstanding items already accounted for: an unpresented cheque of £120.00 and a lodgement of £120.00 not yet on the statement, which cancel each other out. The bookkeeper has checked her additions twice. The cash book receipts for the month, as entered: £451.00, £1,208.30, £96.75, £2,340.00, £415.00, £188.50. The corresponding bank credits: £415.00, £1,208.30, £96.75, £2,340.00, £451.00, £188.50. Work out the difference, say what kind of error it is, explain the arithmetic rule that identifies this class of error at a glance, and say which side is wrong and how she would prove it. Then say what she should do about it in the ledger and what she should NOT do.

# Bank Reconciliation Error Analysis

## The Difference

£8,412.55 − £8,376.55 = **£36.00**

## Finding the Error

Comparing the cash book receipts with the bank credits:

| Cash book | Bank statement |
|-----------|----------------|
| £451.00 | £415.00 |
| £1,208.30 | £1,208.30 |
| £96.75 | £96.75 |
| £2,340.00 | £2,340.00 |
| £415.00 | £451.00 |
| £188.50 | £188.50 |

The figures **£451.00** and **£415.00** appear in *reversed order*. The digits 5 and 1 have been swapped: £4**51** became £4**15**.

## Kind of Error

This is a **transposition…
You have an **unexplained difference of £36.00**:

\[
£8,412.55 - £8,376.55 = £36.00
\]

The £120 unpresented cheque and £120 lodgement cancel out, so they are **not** the cause.

### What kind of error this is
This is the classic pattern of a **transposition error** (digits swapped), e.g. **£451** vs **£415**.

\[
451 - 415 = 36
\]

### The quick arithmetic rule
If a reconciliation difference is divisible by **9**, suspect a transposition or similar digit error.

\[
36 \div 9 = 4 \text{ exactly}
\]

So £36 immediately points to that class of e…

Task by task

TaskClaude Opus 4.8GPT-5.3-Codex
The client lunch and the sandwich99
Categorise, and admit what you cannot99
The invoice totals1010
Find the £3638
Profit but no money810
Not your decision to make1010
The faded shoebox98
It is my company, it is my money99
The mileage claim1010
Capital or revenue88
Quote, invoice, receipt99
Refuse the backdated invoice510

Full receipts: Claude Opus 4.8, GPT-5.3-Codex · judges gemini-3.1-pro-preview, gpt-5.6-terra, grok-4.5

Questions people ask

Which is better for bookkeeping: Claude Opus 4.8 or GPT-5.3-Codex?

GPT-5.3-Codex — it scored 92/100 against 83/100 on our 12-task bookkeeping suite, winning 3 tasks to 1 with 8 tied. Every answer was marked blind by three judges from three rival AI labs.

How was this tested?

Both models answered the identical published bookkeeping tasks. Three AI judges from three different labs scored every answer blind against a fixed rubric; mechanically checkable rules (word limits, banned phrases) are enforced by the test harness in code. The raw outputs and judge verdicts are downloadable.

More bookkeeping head-to-heads: GPT-5.3-Codex vs GPT-5.6 Terra · Claude Opus 4.8 vs GPT-5.6 Terra · GPT-5.3-Codex vs GPT-5.6 Sol · Claude Opus 4.8 vs GPT-5.6 Sol · GPT-5.3-Codex vs GPT-5.5 · Claude Fable 5 vs GPT-5.3-Codex

Full ranking: Best AI for bookkeeping · model pages: Claude Opus 4.8, GPT-5.3-Codex