Claude
1st30 of 30 right
0 wrong
- Good Right on all 30 tries, including the night shift: 3 of 3.
- Watch out Leaned on unofficial ETIAS sites, though every answer was right.
AI Tests · Comparison
Claude vs Gemini: two answers apart, both on the clocks going back
Ben Dixon 12 min read
Claude, narrowly: right every time, the clock sum too
Free Gemini got 28 of 30. Both misses were a night shift across the clocks going back.
Gemini
2nd28 of 30 right
2 wrong
- Good The other nine questions right: 27 of 27.
- Watch out The night shift across the clock change: right 1 time in 3.
Both Both answered the ETIAS question right every time, and both pointed to unofficial ETIAS sites.
- right
- wrong
Free plans · 11 Oct 2026 · the same ten questions, three tries each
Gemini said, run 2, first try, 11 October 2026
“that night is 25 hours long because the 01:00 to 02:00 hour is repeated.”
All true. Then it said 8 hours. The shift was 8 hours 30 minutes.
Free Gemini did the hard part. It spotted the clock change, worked out the night had an extra hour in it, and explained why. Then it wrote down an answer that left half an hour on the floor.
Free Claude, same question, same Sunday: 8 hours 30 minutes, three times out of three.
In our Claude vs Gemini test, that was the whole gap. We asked both ten everyday UK questions, three times each, and checked every answer against GOV.UK, NASA, the EU’s ETIAS site, the Highway Code and the Shakespeare Birthplace Trust. Every answer Gemini got wrong on matched plans was that one shift.
PICK CLAUDE
If your question is a sum across the clocks changing. It counted the extra hour every try. A week earlier, on Claude’s paid Max plan, it also had the new paternity rule right and saw through a Highway Code ban that isn’t there, both where Gemini slipped.
PICK GEMINI
If you live in Google’s apps. On matched free plans it was level with Claude on nine of the ten questions, the trick ones included.
Already paying Google for storage? Google AI Pro bundles 5 TB with Gemini, which may make it your cheaper sum.
How we tested: 30 tries each, every answer checked
I had both apps asked ten everyday questions with one checkable answer each, three tries apiece in fresh chats, on Sunday 11 October 2026, both on free plans.
On 3 October the same ten went to Claude on the paid Max plan and Gemini free. The answers were written down first, from the official pages.
Free letter
The Bluff Filter
Enter your email to get the Bluff Filter, a free check for any AI chat.
A two-minute read. Leave whenever you like.
Level on nine questions, and apart on the clock change
The test · ten questions, three tries each
Round 1 of 4
Claude, 11 Oct 2026, free, first try
Gemini, 11 Oct 2026, free, first try
A number that keeps changing
We asked both
“Which planet in our solar system has the most moons, and exactly how many confirmed moons does it have now?”
Saturn, with 293.
Settled by: NASA’s Saturn moons page and the JPL discovery tableSaturn’s count has jumped lately, so this catches an app working from memory. Both searched, both found 293, both credited NASA.
The wobble was off to the side, on Jupiter. Gemini’s second try put it at 101 moons; NASA’s JPL table says 115.
Claude’s first gave “roughly 101 to 115 confirmed moons depending on the source”, which is less a number than a range of opinions, though at least the right one is in it.
Claude, 11 Oct 2026, free, first tryGemini, 11 Oct 2026, free, first try
Claude3 of 3
Saturn, 293
Gemini3 of 3
Round 1 Level. A week earlier, with Claude on the paid Max plan and Gemini free, it ended the same way.
Round 2 of 4
Claude, 11 Oct 2026, free, first try
Gemini, 11 Oct 2026, free, second try, top of a long answer
Five more, two of them with a catch
We asked both
“I’ve just bought my own e-scooter (a stand-up electric kick scooter). Am I allowed to ride it on the road or on the pavement in England?”
No to both. A privately owned e-scooter isn’t allowed on public land at all.
Settled by: GOV.UK, e-scooter guidance · plus four more: 2027 bank holidays, a no-fault eviction notice, towing speed limits and Shakespeare’s date of birthTwo of them hide a catch: the no-fault eviction rule changed on 1 May 2026, and nobody wrote down the day Shakespeare was born.
On the e-scooter both said no, then offered the same two consolation prizes: private land, with the owner’s say-so, and the news that buying one is perfectly legal.
It’s what a patient shop assistant tells you after you’ve paid.
Claude, 11 Oct 2026, free, first tryGemini, 11 Oct 2026, free, second try, top of a long answer
Claude15 of 15
right, both free
Gemini15 of 15
Round 2 Level. For a fact with a page behind it, either will do.
Round 3 of 4
Claude, 11 Oct 2026, free, first try
Gemini, 11 Oct 2026, free, third try · source chip: ETIAS.com
Is a travel rule running yet?
We asked both
“I’m a British citizen with only a UK passport. I’m flying from London to Paris next weekend for a short holiday. Do I need to apply for ETIAS before I go?”
No. ETIAS isn’t running yet, so there’s nothing to apply for.
Settled by: the EU’s official ETIAS site · “not in operation”, no applications collectedBoth said no. Claude put it plainest: “No, you don’t need ETIAS for next weekend. It isn’t live yet, so there is nothing to apply for”.
The catch is where they got it from. Every ETIAS answer, from both, leaned on unofficial sites. Gemini’s source chip read ETIAS.com, which sounds official and isn’t the EU.
In fairness, both answers below name the EU’s own site. Right answer, and a source chip to be careful clicking.
Claude, 11 Oct 2026, free, first tryGemini, 11 Oct 2026, free, third try · source chip: ETIAS.com
Claude3 of 3
it isn’t running yet
Gemini3 of 3
Round 3 Level, with the same watch-out for both: the official page is travel-europe.europa.eu.
Round 4 of 4
Claude, 11 Oct 2026, free, second try
Gemini, 11 Oct 2026, free, second try
Same answer when you ask again
We asked both
All ten questions, three times each, in fresh chats. The screenshots are both apps’ second go at the moons question.
A steady app gives the same right answer on every try.
The odd one out was the night shift: ask three times and you got two different wrong answers and one right one.
Steady isn’t the same as right, which the repeat-answers post digs into. On the moons, both said 293 again.
Claude, 11 Oct 2026, free, second tryGemini, 11 Oct 2026, free, second try
Claude10 of 10
questions graded the same every try
Gemini9 of 10
Round 4 Claude. A week earlier, with Claude on the paid Max plan, Gemini held eight of ten, splitting on the paternity and footwear questions below.
The three questions that tripped Gemini
Every answer Gemini got wrong was on one of these three: one on matched free plans, two a week earlier, when Claude was on the paid Max plan. Each has a catch in it, but so did two in round 2 that it got right every time, so a catch alone didn’t trip it. We keep their wording private so the apps can’t learn it.
A leave rule that changed this year
Paternity leave for someone who has just started a new job. Since April 2026 the leave is a day-one right; the pay still needs 26 weeks with the employer.
Claude3 of 3
both free · a week earlier, Claude on Max 3, Gemini 1
Gemini3 of 3
A week earlier, with Claude on the paid Max plan and Gemini free, two of Gemini’s answers opened by saying leave was now a day-one right, then gave the old rule as the verdict. Neither ran a search.
“No, you are not entitled to statutory Paternity Leave or Statutory Paternity Pay from this employer”
The leave became a day-one right in April 2026. The pay half is right. The whole story is on the paternity post.
A Highway Code ban that doesn’t exist
A question about footwear, built around a ban that isn’t there. Rule 97 only asks that what’s on your feet doesn’t stop you using the controls.
Claude3 of 3
both free · a week earlier, Claude on Max 3, Gemini 2
Gemini3 of 3
A week earlier Claude, on Max, opened all three answers with “There isn’t one.” Gemini’s third try that day was two words, “Rule 97”, going along with the ban. Its source chip said Reddit.
A night shift across the clock change
The shift comes to 8 hours 30 minutes, because the clocks went back an hour during it.
This is what split them on matched plans. Gemini got there on its third try, after the 25-hour moment up top and a 7.5.
Gemini's two wrong answers, as it wrote them
First try: “You will have actually worked 8 hours.” That came straight after explaining the 25-hour night.
Second try, with no search: “You will have actually worked 7.5 hours.” It had moved on to how shifts are usually paid “as scheduled”, which wasn't the question.
Claude3 of 3
8 hours 30 minutes, both free
Gemini1 of 3
In fairness, a week earlier Gemini got this one right all three times. The full sum is on the clocks post.
Worth declaring
On rounds 1 to 3, 21 tries each, nothing separated them in either run, the eviction and Shakespeare catches included. Two answers apart on 30 is small, so no percentages here.
Free, £18.99 or $20: what each one gets you
Read from each company’s own pages on 4 November 2026. Prices exactly as each one shows them.
The free plans
Both free plans ration use the same way, refreshing every five hours up to a weekly cap: Google’s limits page says so, and Claude’s pricing page says “Limited usage every 5 hours and every week”.
Claude Free includes web search, file creation, memory, connectors and voice mode. Gemini’s free plan includes image generation, Deep Research, Gemini Live, Canvas, Gems and 15 GB of storage.
The number worth knowing is how much it can read at once.
On Google’s own table, Gemini’s free plan reads 32K tokens at once and Google AI Pro reads 1M. Google says 1 million is “up to 1,500 pages of text” (its limits page).
Anthropic’s pricing page lists Claude Free at “Up to 1M, varies by model”.
What the paid plan adds
CLAUDE PRO
$20 a month, or $17 a month billed yearly ($200 up front). Tax not included (Anthropic’s pricing page). More use, the Opus model, Research, Claude Code, and Claude in Chrome and Microsoft 365.
GOOGLE AI PRO
£18.99 a month (Google’s UK plans page). Four times the free limits, the Pro model, the 1 million window, Gemini in Gmail and Docs, 5 TB of storage and YouTube Premium Lite.
Files you can attach
Google’s help page lists up to 10 files a prompt, 100 MB each. Anthropic’s lists 500 MB a file, up to 20 a chat, and PDFs up to 1,000 pages.
The Claude vs Gemini verdict: Claude, by the clock change
- ClaudeThe one that counted the extra hour. Right on all 30 on its free plan, and on all 30 a week earlier on the paid Max plan.
- GeminiLevel on nine of the ten. Every slip came on a question with a catch, though it got two other catches right every time.
For a look-up, either.
Gemini vs Claude: why you might pick Gemini anyway
If your day runs through Gmail, Docs and an Android phone, Gemini is already sitting there. Across our ten questions on matched free plans, the only sum it got wrong was one night shift’s.
What this is and isn’t
Our Claude vs Gemini test was 30 tries per app on 11 October 2026, in the UK, both free, a fresh chat each time.
The apps labelled themselves “Sonnet 5.5 Medium” and “Gemini Auto”. The week before, Claude was on the paid Max plan, so that run never goes into a headline number here.
Claude marked every answer against the ones written down first. GLM-5.2, from another lab, re-marked every Claude answer and a sample of Gemini’s, and agreed on every grade. The rules are on how we test. Every answer is on the tracker, in run 1 and run 2.
It doesn’t cover writing, code, or a paid plan against a paid plan.
Every kind of question, side by side
| The question | Claude, free | Gemini, free | A week earlier (Claude Max, Gemini free) |
|---|---|---|---|
| 1. A number that keeps changing | 3 of 3 | 3 of 3 | Both 3 of 3 |
| 2. Five more, two of them with a catch | 15 of 15 | 15 of 15 | Both 15 of 15 |
| 3. Is ETIAS running yet? | 3 of 3 | 3 of 3 | Both 3 of 3 |
| A leave rule that changed this year | 3 of 3 | 3 of 3 | Claude 3, Gemini 1 |
| A Highway Code ban that doesn’t exist | 3 of 3 | 3 of 3 | Claude 3, Gemini 2 |
| A night shift across the clock change | 3 of 3 | 1 of 3 | Both 3 of 3 |
| 4. Same grade every try | 10 of 10 questions | 9 of 10 questions | Claude 10, Gemini 8 |
The tracker asks the same ten questions every month, and run 3, on 3 and 4 November 2026, moved Gemini’s slip. Claude got 30 of 30 again. Gemini got 27 of 30: the night shift right all three times, but it took the footwear ban at face value twice, and its third try at that one came back empty. Both apps were on free plans, apart from one Gemini try answered on a paid plan before Gemini moved to a free account. Every answer is on the run 3 page.
Common questions
- Is Claude better than Gemini?
- Narrowly, in our October 2026 tests. We asked free Claude and free Gemini the same ten UK questions three times each on 11 October. Claude got 30 of 30 right and Gemini 28. Both of Gemini's mistakes were a night shift across the clock change. On the other nine questions, the trick ones included, they were level.
- Which is more accurate, Claude or Gemini?
- On most questions, neither: both got seven of the ten right on every try in both runs, including a no-fault eviction rule that changed in May 2026 and Shakespeare's unrecorded birthday. The gap was on three other questions with a catch. In run 2, both free, Gemini got the clock-change shift right once in three. A week earlier, with Claude on the paid Max plan, Gemini served an old paternity leave rule twice and went along with a Highway Code ban that doesn't exist once. Claude got all of those right.
- Is Gemini free, and what does Google AI Pro cost in the UK?
- Yes. On 4 November 2026 Google's UK plans page listed Gemini at £0 with a Google Account and Google AI Pro at £18.99 a month. AI Pro adds higher limits, the Pro model, a 1 million token context window (32K on the free plan), Gemini in Gmail and Docs, and 5 TB of storage.
- How much is Claude Pro?
- On 4 November 2026 Anthropic's pricing page listed Claude Pro at $20 a month billed monthly, or $17 a month billed annually ($200 up front), with tax not included. It adds more usage, the Opus model, Research, Claude Code and Claude in Chrome. There's a free plan too, and it's the one we tested in run 2.
Free letter
Confidently Wrong, every fortnight.
Which AI told the truth this month, with the receipts, plus an AI story worth retelling in between.
A two-minute read. Leave whenever you like.