GLM 5.2 is nearly as accurate as a human book keeper

The real test to see if AI is just a rich person thing will be to see how the tax authorities treat it, even for more complex returns.

They can save humans for the really complex edge case stuff but at the end of the day, the tax code is just checkboxes and input forms that get boiled down into Integers, Floats/Doubles and enumerated choices with some Strings for deductions

I just have a folder on my computer where I keep things in beancount. Then I have mercury CLI access with a read token to my business bank account, and I have my emails fully synced in there as well via IMAP. Claude Code with Opus just seamlessly hooks everything up so my accounts are up to date. At the end of the year, I used that information to prepare my tax returns for the business and then later the part that flowed to me as the owner.

I had a fairly complicated tax return in 2025 involving a couple of change of business tax consideration and some money that was accidentally sent to me as a 1099 instead of to the business and I did everything with tax software with Claude Code advising.

The end result was pretty damned good. I was unsurprisingly audited (or at least carefully reviewed) and the only error was in some way where I allocated a small amount of my wife's tax-free disability payments (disability is the mechanism that California uses to provide maternal benefits pay protection; she's not actually disabled). The IRS told me about it, I paid that bit (it was meant to be claimed back from the employer, not the US government) and everything was hunky dory. To be honest, the sum was so small I did not investigate (and haven't yet followed up with getting reimbursed by her employer).

Honestly, almost all of it could have been avoided if I'd paid an accountant and a tax lawyer and they'd told me things and I'd done as they did, but in the end the combination of the fact that the IRS is very reasonable when you explain things and a modern agent means that the entire process was quite simple. In the end, I preferred the interactive mechanism of working with software because most accountants and lawyers will prefer to get all of your documentation all at once and then work on it rather than do it incrementally. In my case, I was able to work on the return incrementally and then have everything plugged in. I could ask a bunch of questions and get clarification.

I think I will probably do all this the same way this year (though of course my taxes will be simpler).

Thanks, that's useful! We did use ChatGPT 5.5 for some time and it did perform pretty well too (of course more expensive than GLM 5.2). We tried Claude 4.7 and 4.8, but we found both models to be "lazy" and very expensive. Claude would always rather prefer the route of saying that the evidence was not found or something is incomplete, rather than put more effort into finding/repairing the particular issue.

I've submitted my German taxes this year using a mix of Claude 4.6 and Claude 4.7, with lots of manual checking. The German Finanzamt granted most of the things I listed in the tax return (they send you an official letter by post) -- I did have to appeal for one of the items though (again using Claude, this time 4.8 ).

The most important thing I've found is to ask Claude to thoroughly audit the reply (to find all hallucinations). I usually ask it to give me an enumerated list of all facts and all legal cases quoted, and then I give it to a new instance to carefully validate each one.

Newer models are getting much better at not hallucinating German case law though :)

Why do you think you were audited?