Grok vs ChatGPT: ChatGPT answered all 90. Grok answered 33

pass · partial · miss · confidently wrong

ETIAS, the EU’s travel permit, isn’t running yet, so a British passport holder has nothing to apply for. Grok opened all three of its answers with yes. Run 1’s then said plainly this trip didn’t need it, and was marked right.

Run 2’s two kept it up front, “Yes, you should apply for ETIAS well in advance of your trip”, and walked it back further down. Neither labelled its answer, so we mark the one you meet first.

It’s a waiter recommending the special, then mentioning the kitchen’s shut.

Shakespeare went Grok’s way in October. Both apps opened on 23 April 1564. All three times it answered, Grok added that no surviving document records the birth and that he was baptised on 26 April 1564, as the Shakespeare Birthplace Trust says. In November ChatGPT got all three right, and Grok gave no answer.

One was a Highway Code question about footwear, built around a ban that doesn’t exist. ChatGPT’s whole answer was the rule number, “97” or “Rule 97.”, every time. Grok’s two replies named a rule too, and in run 2 it was “Rule 248.”, which is about parking at night.

Rule 97 only asks that your footwear doesn’t stop you using the controls, and neither app said there’s no such ban. ChatGPT’s side of it is on is ChatGPT accurate.

The other was a night shift across the October 2026 clock change, which comes to 8 hours 30 minutes. ChatGPT mostly landed on 7.5.

Grok missed every time, and in run 2 it opened on 22 hours 30 minutes, then 10 hours, then 27.5. By the third go the shift was longer than a day, which would at least explain feeling tired.

Run by run, ChatGPT got 22, 23, then 24 right of 30. Grok got 8, 6, then 6, of the 12, 12 and 9 it answered. The other four apps got the same questions and answered 359 of their 360 tries, on run 1, run 2 and run 3 of the tracker.

I had this run through an automated browser in the UK, waiting up to 12 minutes a reply, which is longer than anyone would. ChatGPT was signed out in run 1, then free on its “Think” setting (on my own free account in run 3), and Grok free on “Fast” every time. Memory was off on both, and web search left as each app came.

Of Grok’s 57 empty tries, 49 had already had a second go in a fresh chat. The other 8 came on 4 November, when free Grok sent strings of unconnected words in place of answers. It still did at our last check that night, with no second go possible before our two-day window to ask again closed. We wrote every answer key before the first question went out. Claude marked every answer, and GLM-5.2, a model from another lab, marked a sample again. The rules are on how we test.

The test can’t say why Grok’s replies stopped or whether paid plans behave differently, and we didn’t test coding, images, writing or posts on X.