ChatGPT vs Claude vs Grok: we tested them on 5 everyday tasks

01How we ran it (and what is not equal)

• Grok: run by our agent Grok Bot on grok.com, in the mode Grok shows as Fast, on October 19, 2026.
• Claude: Sonnet 5.5, which Claude's free plan includes, in a fresh chat with no tools and no memory, on October 21, 2026.
• ChatGPT's model: OpenAI's GPT-6.1 through OpenAI's Codex tool at its "ultra" reasoning setting, on October 21, 2026. That is not the ChatGPT app, and free ChatGPT accounts get a different default model.

One run per prompt, so this is a test, not a study. Every answer below is real and unedited.

02The four checks

1. Did it do exactly what I asked? The number, the limit, the format.
2. Do the numbers add up? We redid the math inside each answer.
3. Did it invent anything? No new facts about your situation.
4. Could I use it today without fixing it?

03Task 1: the $600 trip is where the math broke

Plan a 3-day trip to Quebec City for two people with a $600 budget, day by day.

Grok's own table totals $450 to $620, so its top end is $20 over budget. It also calls the Lévis ferry "free", then prices it at about $8 to $9 round trip for two, while its own table says about $4.25 one way. Claude's table adds up to exactly $600, but its Day 2 says "about $110" while the items listed that day add up to $155 to $165:

**Day 2: History, parks and a nicer dinner (about $110, including one paid attraction)**
- Morning: Walk the top of the old city walls (free), then go to the Plains of Abraham (free), which has big views of the St. Lawrence.
- Optional paid pick, about $40 for two: a Citadelle tour or the Musée de la civilisation. Pick one, or skip it and keep the money.
- Lunch: The Marché du Vieux-Port (Old Port Market), where you can graze on local cheese, bread and fruit for about $25.
- Afternoon: Walk rue Saint-Jean and then the Saint-Jean-Baptiste and Saint-Roch neighbourhoods, with cafés, street art and shops.
- Evening: This is your splurge. Have a bistro dinner for about $70 to $80 for two, then drinks at a local microbrewery such as La Barberie (about $20).

GPT-6.1's plan is plainer, with fewer named places, but every number checks: $210 + $210 + $70 for the three days, plus a $110 cushion, makes $600.

04Task 3: the rewrite that invented a detail

Rewrite this so it is polite but firm: "The heating has been broken for a week and nobody came."

Hello,

The heating has been out of order for a week now, and no one has come to fix it despite my earlier requests. Please arrange for a technician to come as soon as possible and let me know the expected time.

Thank you.

"Despite my earlier requests" is not in the original sentence. If you never asked before, that line can backfire. Claude and GPT-6.1 stuck to the facts. GPT-6.1 wrote the shortest one, two sentences:

The heating has been out of service for a week, and no one has come to repair it. Please arrange an urgent repair and confirm today when a technician will arrive.

05Tasks 2, 4 and 5: close, with one trap each

Explain how a heat pump works to a 12-year-old in 5 sentences.

All three explained the heat pump in exactly five correct sentences: 4/4 each.

Give me a 7-day plan to learn 100 Spanish words in 15 minutes a day.

Claude (6 days of 16 to 17 words) and GPT-6.1 (5 days of 20 words) both list exactly 100 words with no repeats. Grok's Day 7 promises 12 new words and lists 13.

I have chicken, rice, spinach and lemons. Give me 3 dinners under 30 minutes, with steps.

Nobody got full marks here. Grok gave the same lemon chicken, spinach and rice plate three ways. Claude gave three different dinners, including a soup, but two of them say "about 30 min", not under 30. GPT-6.1 kept all three at 25 minutes only by switching to quick-cooking rice, and said so in bold.

06Verdict

For everyday tasks, all three are good enough. GPT-6.1 won on points (19/20) with the plainest answers. Claude (18/20) wrote the richest trip plan and the most varied dinners. Grok (15/20) slipped on fine print. Settings matter: we ran OpenAI's model at its "ultra" reasoning setting and Grok in Fast mode, so treat this as a snapshot. Whatever you use, run the four checks before you book, send or cook.