Battles / Everyday chat
Too close to call
ChatGPT vs Gemini · Everyday chat
The gap between them is smaller than this test can reliably measure, so we will not name a winner on quality. The full statistics are below.
ChatGPT 3, Gemini 1. That margin is inside the noise of our suite, so we do not call a winner on quality. The judge could pick a winner on 4 of 18 tasks; on the other 14 it could not tell them apart.
But ChatGPT answered noticeably faster.
Show the full workings
The evidence
Suite-by-suite
blue = ChatGPT wins · grey = ties · white = Gemini wins (18 tasks per suite)
Round-by-round — all 18 tasks
tieExplain compound interest · 9.2 v 9Both are accurate, concise, within word limit, avoid banned words, and use the same clear worked example. Response 2 has slightly friendlier tone…
Explain compound interest to a 15-year-old in no more than 120 words. Include one worked example with real numbers. Do not use the words 'exponential' or 'snowball'.
Compound interest means you earn interest on both your original money and the interest already added. Example: You save **$100** in an account paying **10% interest each year**. - After year 1: $100 + $10 interest = **$110** - After year 2: 10% of $110 = $11 interest, so you have **$121** - After year 3: 10% of $121 = $12.10 interest, so you have **$133.10** Because the interest is added to your balance each year, later interest payments become larger.
Compound interest is when you earn money on your original savings, plus on the interest those savings already earned. It’s basically getting paid interest on your interest! Imagine you put $100 in an account earning 10% interest each year. After Year 1, you earn $10. You now have $110. In Year 2, you earn 10% on that new $110 total. That’s $11, giving you $121. In Year 3, you earn 10% on $121, which is $12.10. Now you have $133.10. Even though you never added more of your own cash, the free money you receive grows every single year because your total balance keeps getting bigger.
anthropic/claude-sonnet-5: tie (flipped on swap) — Both are accurate, concise, within word limit, avoid banned words, and use the same clear worked example. Response 2 has slightly friendlier tone but both are essentially equally effective for a teen audience. / Both are accurate, within word limit, avoid banned words, and include clear examples. Response 2 is slightly more concise and well-formatted with bullet points, aiding readability for a non-technical teen, giving it a slight edge. | x-ai/grok-4.5: tie — Both accurate, under 120 words, include clear numbered example, avoid banned terms, suitable for teens, concise and useful. / Both accurate, under 120 words, include clear numbered example, avoid banned terms, suitable for teen. Equally clear and useful. | deepseek/deepseek-v4-pro: tie — Both accurately explain compound interest with a correct worked example, stay within 120 words, and avoid forbidden words. Response 1 is slightly more structured with bullet points, while Response 2 is more conversational. Both are equally clear and useful for a 15-year-old. / Both responses correctly explain compound interest with a clear worked example, stay within the word limit, avoid banned words, and are equally useful and concise.
tiePhone contract maths · 10 v 10Both responses are correct, clear, well-structured, and arrive at the same accurate arithmetic and conclusion. Equally useful and concise for a…
Which is cheaper over three years: (a) a £35/month phone contract with a free handset, or (b) buying the handset outright for £300 plus a £12/month SIM-only plan? Show the arithmetic for both and state the winner and the difference.
Over 3 years = 36 months. **(a) £35/month contract with free handset** £35 × 36 = **£1,260** **(b) £300 handset + £12/month SIM-only** Handset: £300 SIM-only: £12 × 36 = £432 Total: £300 + £432 = **£732** **Winner:** Option **(b)** — buy the handset outright and use the SIM-only plan. **Difference:** £1,260 − £732 = **£528 cheaper** over three years.
First, let's establish the timeframe: 3 years is equal to **36 months** (3 x 12). Here is the arithmetic for both options: **Option (a): £35/month contract with a free handset** * Monthly cost: £35 x 36 months = £1,260 * Upfront cost: £0 * **Total cost: £1,260** **Option (b): Buying the handset outright plus a £12/month SIM-only plan** * Monthly cost: £12 x 36 months = £432 * Upfront cost: £300 * **Total cost:** £432 + £300 = **£732** **The Winner and Difference:** * **Winner:** Option (b) is cheaper. * **Difference:** You save **£528** over three years by choosing option (b) (£1,260 - £732 = £528).
anthropic/claude-sonnet-5: tie — Both responses are correct, clear, well-structured, and arrive at the same accurate arithmetic and conclusion. Equally useful and concise for a non-technical reader. / Both responses are correct, clear, and identical in arithmetic and conclusion. No constraints violated. Equally useful and concise. | x-ai/grok-4.5: tie — Both correctly compute 36-month totals (£1260 vs £732), show full arithmetic, declare (b) cheaper by £528, and are clear/concise for non-experts. No constraints breached. / Both correctly compute 36 months, totals £1260 vs £732, winner b by £528. Clear arithmetic, fully follow task, equally useful and concise. | deepseek/deepseek-v4-pro: tie — Both responses correctly compute costs, show arithmetic, identify winner (b) and difference (£528), follow all instructions, and are clear and concise. No constraints violated. / Both responses are completely accurate, show all arithmetic, clearly state the winner and difference. No constraints were violated. Both are equally useful and clear for a busy professional.
tieRecipe scaling · 8.2 v 8.3Response 1 is correct and clear: 4 eggs+2 yolks gives 6 yolks/4 whites, reasonable. Response 2's math is confusing and inconsistent (3 eggs+2…
This carbonara serves 4: 320g spaghetti, 2 eggs, 1 egg yolk, 100g pancetta, 50g pecorino. Rescale it for 7 people. Give the new quantities, round sensibly for things you cannot buy in fractions, and say which one you rounded and why.
For 7 people: - **560 g spaghetti** - **4 eggs** - **2 egg yolks** - **175 g pancetta** - **90 g pecorino** The exact scaled amount would be 3½ eggs, 1¾ yolks, and 87.5 g pecorino. I rounded the **eggs to 4 whole eggs and 2 yolks** because eggs can’t practically be bought or used in fractions; this also keeps the carbonara nicely rich. I rounded the pecorino to **90 g** for easier weighing.
To scale the recipe from 4 people to 7 people, we need to multiply all the ingredients by a scaling factor of **1.75** (since 7 ÷ 4 = 1.75). Here are the **new quantities** for 7 people: * **Spaghetti:** 560g *(320g x 1.75)* * **Whole Eggs:** 3 * **Egg Yolks:** 2 * **Pancetta:** 175g *(100g x 1.75)* * **Pecorino:** 88g ### What was rounded and why: **1. The Eggs and Egg Yolks (The main items you cannot buy/use in fractions)** * *The exact math:* 3.5 whole eggs and 1.75 egg yolks. * *Why I rounded:* You cannot buy half an egg, and trying to divide a raw egg white or yolk into fractions is messy, impractical, and unnecessary for home cooking. * *How I rounded:* I rounded to **3 whole eggs and 2 egg yolks**. The exact math calls for a total of 5.25 yolks and 3.5 whites. By using 3 whole eggs and 2 extra yolks, you get exactly 5 yolks and 3 whites. This keeps the rich yolk-to-white ratio of your original recipe almost perfectly intact without wasting partial eggs. *(Note: If you prefer a slightly looser, wetter sauce, you can round the whole eggs up to 4 instead).* **2. The Pecorino Cheese** * *The exact math:* 87.5g. * *Why I rounded:* While you can easily cut or grate cheese to any weight, standard home digital kitchen scales do not measure in half-grams. * *How I rounded:* I rounded it up slightly to **88g** so it is easy to weigh on a standard scale.
anthropic/claude-sonnet-5: tie (flipped on swap) — Response 1 is correct and clear: 4 eggs+2 yolks gives 6 yolks/4 whites, reasonable. Response 2's math is confusing and inconsistent (3 eggs+2 yolks claimed as 5 yolks/3 whites, but original ratio was 3 yolks/2 whites per 4 people, scaled differently), less trustworthy despite detail. / Both scale correctly and round sensibly, explaining rationale. Response 1 gives more precise egg-yolk math (5 yolks/3 whites) matching ratio better, though slightly verbose. Response 2 is more concise but egg rounding is less precise reasoning. Both clear and useful; Response 1 slightly more thorough. | x-ai/grok-4.5: B — Both correctly scale by 1.75 and round eggs/cheese sensibly with reasons. R2 superior on egg ratio preservation and detail; R1 more concise but slightly less precise. / Both correctly scale by 1.75 and explain rounding of eggs/cheese. R1 better preserves yolk ratio and details; R2 slightly less precise but clearer/concise. | deepseek/deepseek-v4-pro: tie (flipped on swap) — Both responses correctly rescale and explain rounding. Response 2's rounding is more precise (3 eggs + 2 yolks yields 5 yolks, closer to 5.25; 88g pecorino vs 87.5g) and its explanation is more detailed, making it more useful. / Both responses are accurate and follow instructions. Response 2 is more concise and its egg rounding (4 eggs + 2 yolks) precisely preserves the original yolk-white ratio, making it slightly more useful for a busy professional.
ChatGPTWifi troubleshooting · 9 v 8Both give concrete, actionable steps within 8 items, ordered by likelihood, and cover Mac/Windows appropriately. Response 1 is more universally…
My laptop will not connect to my home wifi but my phone connects fine. Give me a numbered troubleshooting list, most likely cause first, maximum 8 steps. Each step must be an action I can actually take, not 'check your settings'.
1. **Forget and reconnect to the Wi‑Fi network.** On your laptop, open Wi‑Fi networks, select your home network, choose **Forget**, then reconnect and carefully re-enter the password. 2. **Restart the laptop and your router/modem.** Shut down the laptop. Unplug the router/modem for 30 seconds, plug it back in, wait until Wi‑Fi is fully back, then start the laptop and try again. 3. **Turn Wi‑Fi off and back on, and disable Airplane Mode.** Click the network/Wi‑Fi icon, turn **Airplane mode** off if it is on, turn Wi‑Fi off for 10 seconds, then turn it back on and reconnect. 4. **Move close to the router and try the other Wi‑Fi band.** Stand near the router. If you see separate network names such as `HomeWiFi` and `HomeWiFi-5G` (or `-2.4G`), try the one you were not using. 5. **Run the built-in network troubleshooter.** - **Windows:** Settings → System → Troubleshoot → Other troubleshooters → **Network and Internet** → Run. - **Mac:** Hold **Option** and click the Wi‑Fi icon → **Open Wireless Diagnostics** → Continue. 6. **Disable and re-enable the laptop’s Wi‑Fi adapter.** - **Windows:** Right-click Start → Device Manager → Network adapters → right-click your wireless adapter → **Disable device** → wait 10 seconds → **Enable device**. - **Mac:** System Settings → Network → Wi‑Fi → toggle Wi‑Fi off, wait 10 seconds, then turn it on. 7. **Update or reinstall the Wi‑Fi adapter driver.** - **Windows:** Device Manager → Network adapters → right-click the wireless adapter → **Update driver** → Search automatically. If that fails, choose **Uninstall device**, restart the laptop, and let Windows reinstall it. - **Mac:** Install available macOS updates: System Settings → General → Software Update. 8. **Reset the laptop’s network configuration.** - **Windows:** Settings → Network & internet → Advanced network settings → **Network reset** → Reset now → restart. - **Mac:** System Settings → Network → Wi‑Fi → Details → **Forget This Network**, then reconnect; if needed, remove Wi‑Fi from Network settings and add it again.
Here is a step-by-step troubleshooting list, ordered from the most common and easiest fixes to the more advanced ones. *(Note: These specific actions assume you are using a Windows laptop, which is the most common).* 1. **Restart your laptop.** Click the Start button, click the Power icon, and select "Restart" (do not select "Shut down", as modern laptops often just go into a hibernation state when shut down). 2. **Power-cycle your router.** Unplug the power cable from the back of your home Wi-Fi router, wait exactly 30 seconds, and plug it back in. Wait 3 to 5 minutes for the lights to stabilize, then try connecting your laptop again. *(Routers can occasionally block or glitch out on a single specific device while letting others connect).* 3. **Forget the network and reconnect.** Click the Wi-Fi icon on your taskbar, right-click your home Wi-Fi network name, and select "Forget". Click the network name again, click "Connect", and re-enter your Wi-Fi password. 4. **Toggle Airplane Mode.** Click the Wi-Fi/Network icon on your taskbar, click the "Airplane mode" button to turn it on (this cuts power to the Wi-Fi radio). Wait 10 seconds, then click it again to turn it off and force the laptop to search for signals from scratch. 5. **Hard-reset the Wi-Fi adapter.** Press the **Windows Key + R** on your keyboard, type `ncpa.cpl`, and press Enter. Right-click your Wi-Fi adapter in the window that opens, click "Disable", wait 10 seconds, then right-click it again and click "Enable". 6. **Flush your network cache.** Open the Start menu, type `cmd`, right-click "Command Prompt", and select "Run as administrator". Type `ipconfig /flushdns` and press Enter. Then type `netsh winsock reset` and press Enter. Restart your laptop. 7. **Force a driver reinstall.** Right-click the Start button and select "Device Manager". Click the arrow next to "Network adapters" to expand the list. Right-click your main Wi-Fi adapter (it will usually have "Intel", "Realtek", "Broadcom", or "Wi-Fi" in the name), click "Uninstall device" *(do NOT check the box that says "Attempt to remove the driver for this device")*, and restart your laptop. Windows will automatically reinstall it when it boots up. 8. **Perform a full Windows Network Reset.** Open the Start menu, type `Network Reset`, and press Enter. Click the "Reset now" button and click "Yes" to confirm. Your laptop will wait 5 minutes, restart itself, and wipe all network configurations back to their factory defaults.
anthropic/claude-sonnet-5: A — Both give concrete, actionable steps within 8 items, ordered by likelihood, and cover Mac/Windows appropriately. Response 1 is more universally applicable (Mac+Windows), while Response 2 assumes Windows only, reducing usefulness for non-Windows users despite technical detail. / Both provide concrete, actionable 8-step lists. Response 2 covers both Windows and Mac, adding a band-switching step often overlooked, making it more broadly useful and slightly more concise. Response 1 assumes Windows only but is equally correct and detailed. | x-ai/grok-4.5: A — Both deliver exactly 8 actionable numbered steps starting with likeliest fixes. R1 covers Win/Mac and is slightly clearer/more complete; R2 is Win-only but still highly useful and precise. / Both deliver valid 8-step action lists ordered by likelihood. R2 better covers unknown OS without assumptions, more useful and complete for non-technical users. | deepseek/deepseek-v4-pro: A — Response 1 covers both Windows and Mac, making it more useful for any user. Both follow constraints exactly and are accurate, but 1 is more inclusive and slightly clearer for non-technical professionals. / Both lists are accurate and actionable. Response 2 provides steps for both Windows and Mac, making it more useful for a non-technical professional who may use either OS. Response 1 assumes Windows only, which could be less helpful.
ChatGPTCancel an appointment · 9.3 v 8.3Both meet word limit and requirements; Response 1 is more concise and equally complete, while Response 2 is slightly longer and includes…
Write a message cancelling a dentist appointment two hours beforehand. Apologise, give a brief reason, and ask to rebook next week. Maximum 60 words. Do not grovel.
Hi, I’m sorry for the short notice, but I need to cancel my appointment today due to an unexpected work issue. Could we please rebook for sometime next week? Thank you for understanding.
Hello, I apologize, but I need to cancel my appointment scheduled in two hours due to a sudden illness. I appreciate your understanding regarding the short notice. Could we please rebook for a day next week? Let me know what times you have available. Thank you, [Your Name]
anthropic/claude-sonnet-5: A — Both meet word limit and requirements; Response 1 is more concise and equally complete, while Response 2 is slightly longer and includes unnecessary placeholder text. / Both meet requirements within word limit, apologize briefly, give reason, and ask to rebook without groveling. Response 2 is more concise and natural while still complete, giving it a slight edge in clarity and usefulness. | x-ai/grok-4.5: A — Both meet all requirements with apology, brief reason, rebook request, under 60 words, no groveling. Resp1 more concise and natural. / Both meet all constraints (under 60 words, apologize + reason + rebook, no groveling). R2 is tighter, clearer, and more professional for a busy reader. | deepseek/deepseek-v4-pro: A — Both follow rules, but Response 1 is slightly more concise and directly asks to rebook without asking for available times, which makes it marginally more efficient for a busy professional. / Both meet constraints. Response 2 is more concise and direct, better for a busy professional.
tieVirus vs bacteria · 9.2 v 9Both correct, concise, within word limit, no military analogies, clear explanations. Response 1 slightly more useful with practical note on…
Explain the difference between a virus and a bacterium, and why antibiotics work on one and not the other. Maximum 100 words. Do not use any military or war analogy.
A bacterium is a living single-celled organism that can grow and reproduce on its own. A virus is much smaller and cannot reproduce without entering a host cell and using that cell’s machinery. Antibiotics work against bacteria by disrupting bacterial features such as cell walls, protein-making machinery, or DNA processes. Viruses lack these features and live inside host cells, so antibiotics do not affect them. Viral infections may require antiviral medicines, though many resolve with supportive care.
Bacteria are living, single-celled organisms that reproduce independently and possess distinct features like cell walls. Conversely, viruses are non-living packages of genetic material inside a protein shell; they require a host cell to replicate. Antibiotics function by interfering with biological mechanisms unique to bacteria, such as halting cell wall construction or blocking bacterial metabolism. Because viruses lack these cellular structures and rely entirely on the host's machinery to copy themselves, antibiotics have nothing to interact with. Consequently, antibiotics only affect bacteria and do not work on viruses.
anthropic/claude-sonnet-5: tie — Both correct, concise, within word limit, no military analogies, clear explanations. Response 1 slightly more useful with practical note on antivirals/supportive care; response 2 slightly more detailed on virus structure. Roughly equal quality. / Both accurate, concise, within word limit, no military analogies. Response 2 slightly more informative with antiviral mention; Response 1 slightly more precise wording. Roughly equal quality overall. | x-ai/grok-4.5: tie — Both accurate, under 100 words, no war analogies, clear and useful for non-experts. Equally strong on all criteria. / Both accurate, under 100 words, no war analogies, clear and useful for non-experts. Minor edge to 2 for antiviral note but overall equal. | deepseek/deepseek-v4-pro: tie (flipped on swap) — Both responses are factually correct, follow all instructions (word limit, no forbidden analogies), and are clear and useful for a non-technical professional. No meaningful distinction in quality. / Response 2 adds a practical note on antivirals and supportive care, boosting usefulness for professionals. Both are accurate and follow constraints perfectly. Response 2's extra conciseness and helpful context give it a slight edge.
tieCook from what's in · 5.5 v 6Both add water (necessary but technically not listed), both flag it appropriately. Both provide clear timings/temperatures and stick to listed…
I have chicken thighs, rice, one lemon, garlic and spinach, plus salt, pepper and oil. Give me one dinner recipe using only those. Include timings and temperatures. Do not add an ingredient I did not list.
## One-Pan Lemon Garlic Chicken, Rice & Spinach **Note:** Rice needs water to cook. If water is allowed as a cooking medium, use it as directed below; no other ingredients are added. ### Ingredients - Chicken thighs - Rice - 1 lemon - Garlic - Spinach - Salt - Pepper - Oil - Water ### Method 1. **Heat the oven** to **400°F / 205°C**. 2. **Season the chicken.** Pat the chicken thighs dry. Rub with a little oil, salt, pepper, half the garlic, and finely grated lemon zest. 3. **Brown the chicken.** Heat a large oven-safe skillet or Dutch oven over **medium-high heat**. Add a little oil and place the chicken skin-side down, if it has skin. Cook **6–8 minutes** until browned. Turn and cook **2 minutes** more. Transfer chicken to a plate. 4. **Start the rice.** Reduce heat to **medium**. Add the remaining garlic and cook **30 seconds**. Add the rice and stir for **1 minute** to coat it in the oil and chicken drippings. 5. **Add liquid and bake.** Add water according to your rice’s usual ratio—typically **2 cups water for every 1 cup rice**—plus salt, pepper, and the juice of about half the lemon. Bring to a simmer. Place the chicken thighs on top of the rice. Cover tightly with a lid or foil and bake at **400°F / 205°C for 25 minutes**. 6. **Finish uncovered.** Remove the cover and bake **10–15 minutes more**, until the chicken reaches **165°F / 74°C** in the thickest part and the rice is tender. If the chicken needs more browning, leave it uncovered for the final few minutes. 7. **Add spinach.** Remove from the oven, scatter spinach over the rice, cover again, and let stand **5 minutes** so the spinach wilts. 8. **Serve.** Squeeze over the remaining lemon juice and add more salt and pepper if needed.
Here is a recipe for **One-Pan Lemon Garlic Chicken and Spinach Rice**. Because you do not have chicken broth, we will use the browned bits from the chicken and the garlic to flavor the water used to cook the rice. ### **Ingredients** * Chicken thighs (boneless or bone-in) * 1 cup white rice (rinsed) * 1 lemon (zested and juiced) * 3-4 cloves garlic (minced) * 2 big handfuls of fresh spinach * 2 tbsp oil (divided) * Salt and black pepper * *2 cups tap water (for cooking the rice)* ### **Timings & Temperatures** * **Prep time:** 5 minutes * **Cook time:** 30-35 minutes * **Stove Temperatures:** Medium-High, Medium, High, and Low * **Safe Internal Meat Temperature:** 165°F (74°C) ### **Instructions** **1. Prep the Ingredients** Mince the garlic. Use a grater to remove the yellow zest from the lemon, then cut the lemon in half and squeeze the juice into a small bowl. Rinse your rice under cold water. Generously season both sides of the chicken thighs with salt and pepper. **2. Sear the Chicken (10 minutes)** Place a large skillet (one that has a tight-fitting lid) on the stove and turn the heat to **Medium-High**. Add 1 tablespoon of oil. Once the oil is shimmering and hot, add the chicken thighs. Sear for **4 to 5 minutes per side** until they develop a golden-brown crust. Remove the chicken from the pan and set it aside on a plate. *(Note: The chicken will not be fully cooked yet).* **3. Toast the Rice and Garlic (2 minutes)** Turn the stove heat down to **Medium**. Add the remaining 1 tablespoon of oil to the skillet. Add the minced garlic and stir for **30 seconds** until fragrant, scraping up the browned bits of chicken left in the pan. Add the uncooked rice and stir constantly for **1 to 2 minutes** so the grains get coated in the oil and garlic. **4. Simmer the Chicken and Rice (20 minutes)** Pour 2 cups of water into the skillet. Add the lemon juice, half of the lemon zest, 1 teaspoon of salt, and 1/2 teaspoon of pepper. Stir well. Turn the heat to **High** just until the liquid comes to a rolling boil. Once boiling, nestle the partially cooked chicken thighs (and any juices from the plate) back into the rice. Cover the skillet with the lid and immediately drop the heat to **Low**. Let it simmer completely undisturbed for **18 to 20 minutes**. **5. Wilt the Spinach (3 minutes)** After 20 minutes, turn the heat completely **Off**. Remove the lid and check the chicken with a meat thermometer to ensure it has reached an internal temperature of **165°F**. Temporarily move the chicken back to a plate. Immediately dump the fresh spinach into the hot rice. Stir it through the rice for **1 to 2 minutes**; the residual heat will quickly wilt the spinach and the rice will absorb any remaining moisture. **6. Serve** Taste the rice and add a pinch more salt or pepper if needed. Divide the spinach rice onto plates, top with the chicken thighs, and sprinkle the remaining lemon zest over the top for a fresh finish.
anthropic/claude-sonnet-5: tie (flipped on swap) — Both add water (necessary but technically not listed), both flag it appropriately. Both provide clear timings/temperatures and stick to listed ingredients otherwise. Response 2 is slightly more detailed and precise with stove-top method; Response 1 uses oven, equally valid. Quality comparable. / Both use water (implicit, not banned). Response 2 explicitly lists 'Water' as an ingredient, violating the no-added-ingredient rule, while Response 1 only uses water without listing it as a core ingredient, staying closer to constraints. Response 1 is clearer and equally useful. | x-ai/grok-4.5: tie — Both add unlisted water (necessary but violates 'only those'). Else solid one-pan recipes with timings/temps. R2 clearer steps, stovetop only, more precise quantities/usefulness for non-technical user. / Both add unlisted water (necessary but forbidden). Both give valid one-pan recipes with timings/temps using only listed items plus water. Equally clear, useful, accurate. | deepseek/deepseek-v4-pro: tie — Both recipes are accurate, clear, and include required timings/temperatures. However, both add water, an unlisted ingredient, violating the constraint. Otherwise, they are equally useful and well-structured. / Both recipes correctly use only listed items except water, with clear steps, timings, and temperatures. They are equally useful for a busy non-technical user. Water addition violates the ingredient constraint for both.
tieOffside rule · 8.5 v 7.7Both are accurate, within word limit, and clearly state the common misconception. R1 is more precise technically; R2 is more accessible and…
Explain the football offside rule to someone who has never watched a match, in under 80 words. Include the one thing people most often get wrong about it.
A player is offside if, when a teammate plays the ball, they are nearer the opponents’ goal line than both the ball and the second-last opponent, and then become involved in play. No offside applies in their own half, or from throw-ins, corners, or goal kicks. The common mistake: simply standing in an offside position is not a foul—you must affect play.
In football, you can't just wait by the opponent's goal for a pass. When a teammate passes to you, there must be at least two opponents (usually the goalie and one defender) between you and the goal. Otherwise, you're "offside." The biggest misconception? People think offside is judged based on where you are when you *receive* the ball. It’s actually judged at the exact moment your teammate *kicks* it. You can safely sprint past defenders *after* the kick!
anthropic/claude-sonnet-5: tie (flipped on swap) — Both are accurate, within word limit, and clearly state the common misconception. R1 is more precise technically; R2 is more accessible and engaging for a non-technical reader. Both meet all constraints equally well. / Both meet word limit. R2 is more precise and rule-accurate, including key exceptions; R1 is clear and beginner-friendly but slightly oversimplified. R2 edges out for accuracy and completeness. | x-ai/grok-4.5: A — Both under 80 words and cover the rule plus a common error accurately. R1 more precise on involvement and exceptions; R2 clearer on timing myth but slightly looser. / Both under 80 words and cover a key misconception. R2 more factually precise on positioning and involvement. R1 clearer/casual for absolute beginners but slightly imprecise. | deepseek/deepseek-v4-pro: tie (flipped on swap) — Response 1 accurately states the offside rule including the ball condition and exceptions, and correctly identifies the common mistake. Response 2 omits the ball condition, making it less accurate. Both are under 80 words and include the misconception. / Response 1 uses plain, engaging language and a clear example, ideal for a novice. Response 2 is accurate but more technical. Both meet the word limit and include a common mistake.
GeminiPacking list · 7.2 v 9.2Both meet exactly five items with reasons. Response 2 offers more specific, Highlands-relevant, insightful items (deer rutting, OS map, no signal)…
Give me a packing list for a weekend in the Scottish Highlands in October. Exactly five items, no more and no fewer, each with a one-line reason. Assume I already have clothes and a toothbrush.
1. Waterproof hiking boots — Trails and villages can be wet, muddy, and uneven in October. 2. Waterproof jacket — Highland weather can change quickly, with frequent rain and wind. 3. Daypack — Useful for carrying water, snacks, layers, and essentials on outings. 4. Reusable water bottle — Handy for drives, hikes, and avoiding unnecessary plastic. 5. Portable phone charger — Keeps maps, bookings, and your camera available in remote areas.
1. **Headtorch:** Daylight fades quickly in October, and rural Highland paths or roads rarely have streetlights. 2. **Insulated thermos flask:** A hot drink is a morale-saving necessity when facing the inevitably cold and damp Scottish autumn weather. 3. **High-capacity power bank:** Cold temperatures drain batteries rapidly, and you will need your phone for offline maps and photos. 4. **Ordnance Survey (OS) map:** Mobile phone signal is notoriously non-existent in the glens, making a physical map essential for safe navigation. 5. **Binoculars:** October is the peak of the spectacular red deer rutting season, and this allows you to safely observe the wildlife from a distance.
anthropic/claude-sonnet-5: B — Both meet exactly five items with reasons. Response 2 offers more specific, Highlands-relevant, insightful items (deer rutting, OS map, no signal) showing more local knowledge, slightly more useful and vivid than Response 1's generic list. / Both meet the exact five-item constraint with reasons. Response 1 offers more distinctive, Highlands-specific items (OS map, binoculars for rutting deer) demonstrating stronger relevance, while Response 2's boots/jacket arguably overlap with 'clothes' already assumed, slightly violating the spirit of the constraint. | x-ai/grok-4.5: B — Both meet exact 5-item format with reasons. R2 more specific/useful for Oct Highlands (light, maps, wildlife). R1 solid but generic. / R1 gives 5 non-clothing items ideal for Oct Highlands with solid reasons. R2 includes clothes (boots/jacket) despite the assume-already-have note, reducing relevance. | deepseek/deepseek-v4-pro: B — Response 2 includes more location- and season-specific items (headtorch, OS map, binoculars for deer rutting) that better serve a non-technical visitor, while both lists are factually correct and adhere to the five-item limit. / Response 1 is tailored to Scottish Highlands October with specific items and compelling reasons; Response 2 is generic, including likely clothing, making it less useful.
tieChild's question · 9 v 7.8Response 1 is accurate, concise, under 70 words. Response 2 exceeds 70 words (~75), slightly less concise, though also accurate and vivid.
My eight-year-old asks why the sky is blue. Answer the way I should say it to her: under 70 words, scientifically accurate, no 'because of the atmosphere' hand-waving.
Sunlight looks white, but it contains all colors. Air molecules are much smaller than light waves, so they scatter blue light more strongly than red light—like tiny pinballs sending blue light in every direction. When you look up, lots of that scattered blue light reaches your eyes, so the sky looks blue.
Sunlight looks white, but it’s made of all the rainbow colors traveling in waves. Blue light waves are short and choppy. Our air is packed with tiny, invisible gas particles. When sunlight hits them, longer waves pass right through. But the choppy blue waves crash into the particles and scatter all over the sky! That scattered blue light is what you see when you look up.
anthropic/claude-sonnet-5: A — Response 1 is accurate, concise, under 70 words. Response 2 exceeds 70 words (~75), slightly less concise, though also accurate and vivid. / Both are under 70 words, accurate, and avoid vague 'atmosphere' hand-waving. R2 is slightly more precise scientifically (molecule size vs wavelength) and more concise, while R1's 'choppy waves crashing' is a bit less rigorous but still clear for a child. | x-ai/grok-4.5: tie — Both under 70 words, accurate Rayleigh scattering explanations without atmosphere handwaving, clear and kid-appropriate. / Both under 70 words, kid-friendly, accurate Rayleigh scattering explanations without atmosphere handwaving. Equally clear, useful, and compliant. | deepseek/deepseek-v4-pro: tie (flipped on swap) — Response 1 accurately explains Rayleigh scattering with a pinball analogy. Response 2 incorrectly says longer waves 'pass right through'—all scatter, but blue scatters more. Both stay under 70 words and avoid atmosphere hand-waving. Response 1 is more precise. / Both are scientifically accurate, under 70 words, and avoid hand-waving. Response 1 uses more vivid, child-friendly language; Response 2 is slightly more concise. Both equally excellent.
tieLandlord repair request · 9 v 9.2Both meet the 80-word limit and tone requirements. Response 2 is more detailed, professional, and practical (address, access arrangements) while…
Draft a message to my landlord reporting a boiler that has stopped producing hot water, requesting urgent repair. Firm, not rude, and it should make clear this is a heating and hot water issue. Maximum 80 words.
Subject: Urgent Boiler Repair Required Dear [Landlord/Agent], The boiler has stopped producing hot water and is also affecting the heating. This is an urgent issue, particularly as it leaves the property without essential heating and hot water. Please arrange for a qualified engineer to attend and repair it as soon as possible. Please confirm today when this will be addressed. Kind regards, [Your Name]
Dear [Landlord's Name], The boiler at [Property Address] has completely stopped working, leaving us with no hot water and no heating. As this leaves the property without essential services, please treat this as an urgent maintenance issue. I request that you arrange for a heating engineer to repair it as soon as possible. Please let me know immediately when someone will be arriving so I can ensure they have access. Thank you, [Your Name]
anthropic/claude-sonnet-5: tie (flipped on swap) — Both meet the 80-word limit and tone requirements. Response 2 is more detailed, professional, and practical (address, access arrangements) while remaining concise and firm, giving it a slight edge in usefulness and clarity. / Both responses meet the word limit, are firm but polite, and clearly mention heating and hot water issues. Quality and clarity are comparable, with Response 2 having a subject line for clarity, but no significant advantage overall. | x-ai/grok-4.5: tie — Both drafts are accurate, firm yet polite, clearly flag heating plus hot water failure, stay well under 80 words, and suit a busy professional. Minor stylistic differences only. / Both meet the 80-word limit, are firm yet polite, and clearly flag heating plus hot water. Equally clear, useful and accurate. | deepseek/deepseek-v4-pro: tie — Both messages are accurate, firm, polite, and clearly state the heating and hot water issue. Both are within word limit and useful. No clear winner. / Both messages accurately report the boiler issue, request urgent repair, and are firm yet polite. They meet the word limit and are clear and useful for a busy landlord.
ChatGPTRank by footprint · 9.2 v 8.5Both correct, cite Poore & Nemecek, proper ranking. Response 2 is more detailed and useful with explanations, though slightly longer than needed…
Rank these five foods by greenhouse gas emissions per kilogram of product, highest first: beef, cheese, chicken, tofu, lentils. Give an approximate figure in kg CO2e per kg for each and name the type of source that figure comes from.
Approximate global-average greenhouse gas emissions per kilogram of product, highest first: | Rank | Food | Approx. emissions | Source type | |---|---|---:|---| | 1 | **Beef** | **~100 kg CO₂e/kg** | Life-cycle assessment (LCA) meta-analysis | | 2 | **Cheese** | **~24 kg CO₂e/kg** | LCA meta-analysis | | 3 | **Chicken** | **~10 kg CO₂e/kg** | LCA meta-analysis | | 4 | **Tofu** | **~3.2 kg CO₂e/kg** | LCA meta-analysis | | 5 | **Lentils** | **~0.9 kg CO₂e/kg** | LCA meta-analysis | These figures are commonly reported from the global food-system life-cycle assessment dataset of **Poore & Nemecek (Science, 2018)**, as popularized/compiled by **Our World in Data**. They are broad global averages and generally include stages such as farming, land-use change, feed production, processing, transport, retail, and packaging; actual emissions can vary substantially by producer and region.
Here is the ranking of the five foods by greenhouse gas emissions, from highest to lowest. The figures provided are approximate **kilograms of carbon dioxide equivalents per kilogram of product (kg CO₂e/kg)**. **1. Beef** * **Approximate Emissions:** **60 to 100 kg CO₂e/kg** (Emissions vary depending on whether the beef comes from a dedicated beef herd, which is closer to 100 kg, or a dairy herd, which is closer to 33 kg. The global average is roughly 60 kg). * **Why it's high:** Cows are ruminants that produce large amounts of methane (a potent greenhouse gas) during digestion. Additionally, beef requires massive amounts of land and feed, leading to deforestation and fertilizer-related emissions. **2. Cheese** * **Approximate Emissions:** **21 to 24 kg CO₂e/kg** * **Why it's high:** Because cheese is a concentrated dairy product, it takes about 10 liters of milk to make one kilogram of cheese. Therefore, it carries the methane and land-use emissions of dairy cows, plus the energy required for processing and refrigeration. **3. Chicken (Poultry)** * **Approximate Emissions:** **6 to 10 kg CO₂e/kg** * **Why it's lower than beef/cheese:** Chickens are not ruminants, so they do not burp methane. They are also highly efficient at converting feed into meat. The primary emissions come from growing their feed (soy and corn) and the energy used to heat and power poultry facilities. **4. Tofu** * **Approximate Emissions:** **3 to 3.5 kg CO₂e/kg** * **Why it's low:** Tofu is made directly from soybeans. Eating plants directly is vastly more efficient than feeding them to animals. The emissions primarily come from agricultural machinery, fertilizer use, and the energy required to process the beans into tofu. **5. Lentils (and other pulses)** * **Approximate Emissions:** **0.9 to 1.2 kg CO₂e/kg** * **Why it's the lowest:** Lentils are incredibly climate-friendly. As legumes, they fix their own nitrogen from the air, meaning they require little to no synthetic nitrogen fertilizers (which are a major source of agricultural emissions). They also require minimal processing. *** ### Type of Source These figures come from a **meta-analysis of Life Cycle Assessments (LCAs)**. A Life Cycle Assessment is a scientific method used to calculate the environmental impact of a product across its entire lifespan—from land use change and farming (seed to harvest), to processing, transport, retail, and packaging. The specific numbers cited above are drawn from the most comprehensive and widely cited meta-analysis of food LCAs to date: the 2018 study by **Poore and Nemecek published in the journal *Science*** (and subsequently popularized by the data organization *Our World in Data*). This study aggregated data from roughly 38,700 commercial farms in 119 countries. *(Note: While exact numbers can fluctuate based on local farming practices, transportation, and geography, the hierarchical ranking of these specific foods remains consistent across almost all scientific literature).*
anthropic/claude-sonnet-5: tie (flipped on swap) — Both correct, cite Poore & Nemecek, proper ranking. Response 2 is more detailed and useful with explanations, though slightly longer than needed for a busy professional. Response 1 is more concise. Both clear; slight edge to 2 for added context without significant loss of clarity. / Both correctly rank foods with similar figures citing Poore & Nemecek. Response 1 is more detailed but verbose; Response 2 is concise, clear table format, equally accurate and useful. Both meet instructions well. | x-ai/grok-4.5: A — Both factually accurate with correct ranking and Poore/Nemecek LCA source. R1 more concise/table format suits busy pro; R2 adds useful why-details but less tight. / Both correctly rank foods with accurate approx. figures from Poore & Nemecek LCA meta-analysis. R2 more concise/table format better suits busy pro; R1 adds useful but wordier explanations. | deepseek/deepseek-v4-pro: A — Response 1 is more concise and directly answers with a clear table and single figures, while Response 2 adds unnecessary detail for a busy professional. Both are factually correct and follow instructions. / Response 2 is more concise and scannable with a table, directly answering the request without extra explanation, better suiting a busy professional. Response 1 is accurate but verbose. Both follow instructions and are factually correct, though Response 2's beef figure is slightly high for a global average.
tieCouch to 5k · 7.8 v 8.2Both meet exactly 8 lines, no preamble/closing. Response 2 is more concise, consistent format, and clearer progression. Response 1 has slightly…
I cannot run at all and want to run 5k in eight weeks. Give me a week-by-week plan: exactly eight lines, one per week, each stating what I actually do that week. No preamble and no closing paragraph.
Week 1: On 3 nonconsecutive days, walk 5 minutes, then repeat 1 minute easy jog/2 minutes walk 8 times, then walk 5 minutes; walk easily 20–30 minutes on 2 other days. Week 2: On 3 nonconsecutive days, walk 5 minutes, then repeat 1 minute easy jog/90 seconds walk 10 times, then walk 5 minutes; take 2 easy 25–30 minute walks. Week 3: On 3 nonconsecutive days, walk 5 minutes, then repeat 2 minutes easy jog/2 minutes walk 7 times, then walk 5 minutes; take 2 easy walks or rest days. Week 4: On 3 nonconsecutive days, walk 5 minutes, then repeat 3 minutes easy jog/2 minutes walk 6 times, then walk 5 minutes; do one 30-minute easy walk. Week 5: On 3 nonconsecutive days, walk 5 minutes, then repeat 5 minutes easy jog/2 minutes walk 4 times, then walk 5 minutes; do one easy 30-minute walk. Week 6: On 3 nonconsecutive days, walk 5 minutes, then repeat 8 minutes easy jog/2 minutes walk 3 times, then walk 5 minutes; do one easy walk or rest day. Week 7: On 3 nonconsecutive days, walk 5 minutes, jog 12 minutes, walk 2 minutes, jog 12 minutes, then walk 5 minutes; make one session a relaxed 25–30 minute continuous jog if able. Week 8: Do two easy 20–25 minute jogs early in the week with walking breaks as needed, rest or walk lightly for 1–2 days, then complete your 5K at an easy pace using short walk breaks whenever needed.
Week 1: Three days this week, walk 5 minutes, alternate 1 minute of running and 2 minutes of walking for 21 minutes, then walk 5 minutes. Week 2: Three days this week, walk 5 minutes, alternate 2 minutes of running and 2 minutes of walking for 20 minutes, then walk 5 minutes. Week 3: Three days this week, walk 5 minutes, alternate 3 minutes of running and 2 minutes of walking for 20 minutes, then walk 5 minutes. Week 4: Three days this week, walk 5 minutes, alternate 5 minutes of running and 2 minutes of walking for 21 minutes, then walk 5 minutes. Week 5: Three days this week, walk 5 minutes, alternate 8 minutes of running and 2 minutes of walking for 20 minutes, then walk 5 minutes. Week 6: Three days this week, walk 5 minutes, alternate 12 minutes of running and 2 minutes of walking for 28 minutes, then walk 5 minutes. Week 7: Three days this week, walk 5 minutes, run continuously for 22 minutes, then walk 5 minutes. Week 8: Walk 5 minutes and run 25 minutes twice this week, then for your third workout, walk 5 minutes and run your continuous 5k.
anthropic/claude-sonnet-5: B — Both meet exactly 8 lines, no preamble/closing. Response 2 is more concise, consistent format, and clearer progression. Response 1 has slightly convoluted phrasing and extra clauses, less clean structure. / Both give reasonable progressive plans; Response 1 strictly follows exactly eight lines, one per week, no extra info. Response 2 adds extra sentences about walking days beyond core weekly running instructions, slightly cluttering format though still one line per week technically. Response 1 is cleaner and more concise. | x-ai/grok-4.5: tie (flipped on swap) — Both meet exact 8-line format with no extra text. Resp1 more complete progressive plan with recovery days; Resp2 simpler but adequate and clear. / Both deliver exact 8-line plans with no extra text. R1 is cleaner, more progressive and concise for a complete beginner; R2 adds useful extras but is denser. | deepseek/deepseek-v4-pro: A — Response 1 includes rest/walk days and gradual progression, more useful for a beginner. Both follow the 8-line format without preamble. Response 2 is clear but less comprehensive. / Both plans are correct and follow instructions. Response 2 adds rest/walk days, making it more comprehensive and useful for a beginner, while remaining clear.
tieSpot the scam · 8.2 v 9.2Both correctly identify scam signals and give solid advice. Response 2 mixes up 7726 and 60599 reporting numbers (7726 is actually the standard…
I got a text saying 'HMRC: you are due a refund of £284.50. Claim within 24 hours at hmrc-refund-claim.co.uk'. Tell me whether this is a scam, give me the three specific signals in this message that decide it, and tell me what to do next.
This is almost certainly a scam (a “smishing” text). Three specific warning signals: 1. **Fake web address:** `hmrc-refund-claim.co.uk` is not an official HMRC/GOV.UK domain. HMRC services use addresses ending in **`gov.uk`**, such as `www.gov.uk`. 2. **Urgent deadline:** “Claim within 24 hours” is pressure designed to make you act before checking. Genuine tax refunds do not normally depend on clicking a text-link immediately. 3. **Unexpected refund claim by text:** Scammers commonly use a specific refund amount (£284.50) to make the message feel credible. HMRC will not ask you to claim a refund by entering personal or bank details through an unsolicited text link. What to do: - **Do not click the link**, reply, or enter any details. - **Forward the text to 60599** (free in the UK) to report it to your mobile provider, then delete it. - If you want to check whether you are due a refund, go to **https://www.gov.uk/** by typing it into your browser yourself and sign in to your Personal Tax Account—do not use the text’s link. - If you clicked the link or gave details, contact your bank immediately, change any reused passwords, and report fraud to **Action Fraud**.
**Yes, this is absolutely a scam.** This is a classic "smishing" (SMS phishing) attempt designed to steal your personal information and bank details. Here are the **three specific signals** in this message that give it away: 1. **The Fake Web Address (URL):** The link `hmrc-refund-claim.co.uk` is fake. All official UK government websites end in **.gov.uk**. Scammers buy cheap, official-sounding domains (often using hyphens) to trick you into thinking they are legitimate. 2. **Artificial Urgency:** The phrase *"Claim within 24 hours"* is a classic scammer tactic. They create a false sense of panic or urgency so you act quickly without stopping to think critically or verify the information. HMRC will never give you a 24-hour deadline to claim a legitimate tax refund. 3. **The Method of Communication:** HMRC’s official policy states that they will **never** send text messages or emails about tax rebates or refunds that include a link. If you are genuinely owed a refund, HMRC will notify you by post or you will see it when you log into your official Government Gateway account. ### What you should do next: * **Do NOT click the link.** If you haven't clicked it, your phone and data are safe. * **Report the text to your mobile provider:** Forward the message to **7726**. This is a free service in the UK that reports the scam to your mobile network provider so they can block the sender. * **Report it to HMRC:** You can forward the text message to HMRC's official reporting number: **60599** (network charges may apply), or email a screenshot to **phishing@hmrc.gov.uk**. * **Delete and Block:** Once you have reported it, delete the text message and block the number it came from. *Note: If you already clicked the link and entered any banking details, contact your bank immediately using the phone number on the back of your bank card to freeze your account and report the fraud.*
anthropic/claude-sonnet-5: tie (flipped on swap) — Both correctly identify scam signals and give solid advice. Response 2 mixes up 7726 and 60599 reporting numbers (7726 is actually the standard spam-report number, not 60599), causing minor confusion, while Response 1's guidance is consistent and accurate. / Both correctly identify scam, give three accurate signals, and clear actionable next steps. Response 1 slightly more thorough with reporting details (phishing email, 7726). Response 2 slightly cleaner but minor error suggesting forwarding to 60599 rather than 7726 for provider reporting. Both well-formatted, accurate, and useful. | x-ai/grok-4.5: tie (flipped on swap) — Both correctly identify scam, list 3 accurate signals (fake .co.uk, urgency, unsolicited text), and give sound next steps. Minor reporting-number variance; equally clear/useful. / Both correctly identify scam and list 3 signals + next steps. R1 more factually precise on reporting (7726 vs 60599) and HMRC policy. R2 misattributes 60599. No constraints broken. | deepseek/deepseek-v4-pro: B — Response 1 incorrectly says 60599 is for mobile provider, reducing accuracy. Response 2 correctly differentiates provider (7726) and HMRC (60599), uses a clearer structure, and all signals are precise. Thus response 2 is more accurate and useful for a busy professional. / Response 1 is fully accurate, with correct reporting numbers (7726 to mobile provider, 60599 to HMRC). Response 2 incorrectly says forward to 60599 to report to mobile provider, a factual mistake. Winner is Response 1.
tieCompare two decisions · 8.8 v 9Both give nearly identical, correct calculations and reasoning, with same non-financial factor (work-life balance). Response 2 is slightly more…
I can either take a £3,000 pay rise or four extra days of annual leave. I earn £45,000 in England and work a standard 5-day week. Work out which is worth more in cash terms, showing your reasoning, and name one non-financial factor that should also count.
Assuming a standard 5-day week and 52 working weeks: - Working days per year: 5 × 52 = **260 days** - Current daily gross pay: £45,000 ÷ 260 = **£173.08** - Value of four extra paid leave days: 4 × £173.08 = **about £692 gross per year** A £3,000 pay rise is therefore worth much more in cash terms: | Option | Gross annual value | |---|---:| | £3,000 pay rise | **£3,000** | | Four extra annual-leave days | **~£692** | So the pay rise is worth roughly **£2,308 more gross each year**. On a typical England tax setup (standard personal allowance, no student loan or pension deductions), the extra £3,000 would be taxed at 20% income tax plus 8% employee National Insurance, leaving about: - £3,000 × 72% = **~£2,160 take-home** The £692 gross value of the leave is roughly **£498 after equivalent tax/NI**, if you think of it as income you would otherwise earn. One important non-financial factor: **work–life balance and wellbeing**. Four additional days off may be more valuable to you than their salary equivalent if they reduce stress, allow travel or family time, or make the job more sustainable.
To determine which option is worth more in cash terms, we need to calculate the financial value of four working days and compare it to the £3,000 pay rise. Here is the step-by-step reasoning based on standard UK employment and current (2024/25) tax rates for England. ### 1. The Cash Value of Four Days of Leave To find the cash value of four days off, we must calculate your daily rate of pay. * **Working days in a year:** A standard 5-day working week equals 260 working days a year (52 weeks × 5 days). * **Gross daily rate:** £45,000 ÷ 260 = **£173.08 per day**. * **Gross value of 4 days:** £173.08 × 4 = **£692.32**. ### 2. The Cash Value of the Pay Rise The gross value of the pay rise is **£3,000**. Right away, looking at the gross figures, the £3,000 pay rise is worth significantly more than the £692.32 value of the four days of leave. However, to see exactly what goes into your pocket, we should look at the **net (take-home) value**. * **Your Tax Bracket:** In England, a salary of £45,000 puts you in the Basic Rate tax band. A £3,000 raise takes you to £48,000, which keeps you entirely within that same Basic Rate band (the higher rate doesn't start until £50,271). * **Deductions:** You will pay 20% Income Tax and 8% Class 1 National Insurance (the current rate as of April 2024) on this extra money. This is a total deduction of 28%. *(Note: This assumes no student loan or percentage-based pension deductions).* * **Net Pay Rise:** £3,000 minus 28% (£840) leaves you with **£2,160 in cash**. * **Net Value of Leave:** £692.32 minus 28% leaves you with **£498.47 in cash**. **Financial Conclusion:** In strict cash terms, **the £3,000 pay rise is worth significantly more**. Taking the money puts an extra £2,160 in your pocket after tax, whereas the four days of leave are only "worth" about £498 in take-home pay. To make four days of leave financially equal to a £3,000 pay rise, you would need to be earning nearly £195,000 a year. *** ### One Non-Financial Factor to Consider: Work-Life Balance and Burnout While the math heavily favors the money, you should consider your **risk of burnout**. Time is a finite resource. If your current workload is incredibly stressful, or if you frequently find yourself exhausted and lacking time for family, hobbies, or rest, those four extra days (nearly a whole week of paid time off) might have a higher *personal* value to you than the cash. Protecting your mental health and maintaining a sustainable work-life balance can keep you happier and more productive in your career over the long term, which is a benefit a pay rise cannot buy.
anthropic/claude-sonnet-5: tie (flipped on swap) — Both give nearly identical, correct calculations and reasoning, with same non-financial factor (work-life balance). Response 2 is slightly more detailed with tax bracket context and an extra illustrative calculation, but also more verbose. Both accurate and clear. / Both correct and clear; Response 1 gives more thorough reasoning and slightly more detail (e.g., higher-rate threshold check, illustrative extreme salary). Response 2 is more concise but slightly less detailed. Both name work-life balance as non-financial factor. | x-ai/grok-4.5: tie — Both correctly value leave at ~£692 gross vs £3000 rise (net ~£2160), name work-life balance. Clear reasoning, useful, follow all task needs equally well. / Both correctly value leave at ~£692 gross vs £3000 rise (net ~£2160), show clear steps with UK tax, and name work-life balance. Equally accurate, complete, useful and clear. | deepseek/deepseek-v4-pro: tie — Both responses correctly calculate cash values (gross and net) and clearly show the pay rise is worth more. Both name a non-financial factor. Response 1 is slightly more concise with a table; Response 2 is more detailed but equally clear. No constraint violations. / Both responses correctly calculate the gross and net values, show reasoning, and name a non-financial factor. They are factually accurate, follow instructions, and are clear and useful. No constraints were broken.
tieItinerary with constraints · 7.5 v 8.3Both meet constraints, structured clearly with morning/afternoon/evening, avoid queueing, food-focused. Response 1 slightly more concise and…
Plan three days in Rome for a couple who hate queueing and love food. Maximum 200 words, one line per activity, morning/afternoon/evening for each day. Do not recommend anything that normally requires standing in a long line without saying how to avoid it.
Day 1 morning — Espresso and maritozzo at Roscioli Caffè, then wander Campo de’ Fiori and the Jewish Ghetto for carciofi alla giudia. Day 1 afternoon — Take a private/early-entry Vatican Museums tour (prebooked timed ticket) to avoid the usual long queue; lunch nearby at Pizzarium Bonci. Day 1 evening — Aperitivo in Trastevere, followed by a reserved table at Da Enzo al 29 or Cesare al Casaletto. Day 2 morning — Prebook the first-entry Colosseum/Forum/Palatine timed ticket; enter through the designated reservation lane, not the walk-up queue. Day 2 afternoon — Eat supplì at Supplizio, then stroll Monti’s boutiques and have gelato at Fatamorgana. Day 2 evening — Book a food-focused Testaccio walk: Mercato Testaccio tastings, Trapizzino, and dinner at Flavio al Velavevodetto. Day 3 morning — Visit the Pantheon with a timed online ticket, then breakfast pastries at Forno Campo de’ Fiori. Day 3 afternoon — Explore Villa Borghese and book Borghese Gallery tickets well ahead; no walk-up entry is advisable. Day 3 evening — Sunset at Pincio Terrace, then reserve dinner at Armando al Pantheon and finish with gelato at Giolitti.
**Day 1** Morning: Pre-book skip-the-line tickets for the Pantheon, followed by espresso and fresh maritozzi at Roscioli Caffè. Afternoon: Graze on authentic Roman street food among locals at the uncrowded Mercato di Testaccio. Evening: Bypass waits by reserving a table at Flavio al Velavevodetto for legendary cacio e pepe. **Day 2** Morning: Admire art at Galleria Borghese, where mandatory pre-booked timed tickets guarantee zero queueing. Afternoon: Indulge in a private, guided food and wine walking tour through the cobblestone streets of Trastevere. Evening: Relax at Il Goccetto, a cozy, reservation-friendly enoteca serving exceptional regional wines and charcuterie. **Day 3** Morning: Book a 7:30 AM VIP early-access tour of the Vatican to see the Sistine Chapel before the crowds. Afternoon: Master Roman cuisine together during a private, hands-on pasta and tiramisu cooking class. Evening: Enjoy aperitivo at a pre-booked table at The Court rooftop bar, viewing the Colosseum without the entry lines.
anthropic/claude-sonnet-5: tie (flipped on swap) — Both meet constraints, structured clearly with morning/afternoon/evening, avoid queueing, food-focused. Response 1 slightly more concise and word-count safe; Response 2 equally good but slightly less concise. Both under 200 words, no banned queueing without mitigation. / Response 1 fits 200 words and follows structure/constraints well. Response 2 exceeds word count significantly (~230+ words) and has multiple items per slot, violating one-line-per-activity format. | x-ai/grok-4.5: tie (flipped on swap) — Both under 200 words, proper daily M/A/E lines, handle queues via prebooks/reservations, food-focused. R1 slightly tighter format and more varied authentic eats. / Both meet word limit, format, and no-line rules with solid food-focused no-queue plans. Resp1 is cleaner, more precise per-day structure and slightly more useful. | deepseek/deepseek-v4-pro: abstain — judge call failed
tieExplain a bill · 9.3 v 9.3Both correctly compute £97.03 before VAT and £101.88 with VAT, showing clear steps. Response 2 adds a helpful summary and clearer formatting for a…
My electricity bill shows: standing charge 60.10p/day, unit rate 24.50p/kWh, usage 320 kWh over 31 days. Work out the bill before VAT, then with 5% VAT. Show each step.
### 1. Standing charge - Standing charge: **60.10p per day** = **£0.6010 per day** - Number of days: **31** \[ £0.6010 \times 31 = £18.631 \] Rounded to the nearest penny: \[ \boxed{£18.63} \] ### 2. Electricity usage charge - Unit rate: **24.50p per kWh** = **£0.2450 per kWh** - Usage: **320 kWh** \[ £0.2450 \times 320 = £78.40 \] \[ \boxed{£78.40} \] ### 3. Bill before VAT \[ £18.631 + £78.40 = £97.031 \] Rounded to the nearest penny: \[ \boxed{£97.03 \text{ before VAT}} \] ### 4. VAT at 5% \[ £97.031 \times 0.05 = £4.85155 \] Rounded to the nearest penny: \[ \boxed{£4.85 \text{ VAT}} \] ### 5. Total bill including VAT \[ £97.031 + £4.85155 = £101.88255 \] Rounded to the nearest penny: \[ \boxed{£101.88 \text{ including 5\% VAT}} \]
Here is the step-by-step calculation for your electricity bill. *(Note: Energy companies usually calculate the exact amounts in pence and convert to pounds at the end, rounding to the nearest penny. We will do the same here.)* **Step 1: Calculate the total standing charge** Multiply the daily standing charge by the number of days in the billing period. * 60.10p × 31 days = **1,863.10p** (or £18.63) **Step 2: Calculate the total cost of electricity used** Multiply the unit rate by the amount of kWh used. * 24.50p × 320 kWh = **7,840p** (or £78.40) **Step 3: Calculate the total bill before VAT** Add the total standing charge and the total usage cost together. * 1,863.10p + 7,840p = **9,703.10p** * Converted to pounds: **£97.03** (This is your total before VAT) **Step 4: Calculate the 5% VAT** Multiply the total before VAT by 0.05 to find the 5% tax amount. * 9,703.10p × 0.05 = **485.155p** * Converted to pounds and rounded to the nearest penny: **£4.85** **Step 5: Calculate the final bill including VAT** Add the VAT amount to the total before VAT. * £97.03 + £4.85 = **£101.88** **Summary:** * **Total before VAT:** £97.03 * **VAT (5%):** £4.85 * **Total to pay:** **£101.88**
anthropic/claude-sonnet-5: tie — Both correctly compute £97.03 before VAT and £101.88 with VAT, showing clear steps. Response 2 adds a helpful summary and clearer formatting for a non-technical reader; Response 1 uses more math notation. Both accurate and equally useful. / Both correctly compute steps and reach £97.03 before VAT and £101.88 with VAT, showing clear steps. Response 1 is slightly more readable for a non-technical reader; Response 2 uses LaTeX boxes which may render poorly but is equally accurate. | x-ai/grok-4.5: tie — Both correctly compute standing £18.63, usage £78.40, pre-VAT £97.03, VAT £4.85, total £101.88 with clear steps. Equally accurate, compliant, useful and concise for a non-technical user. / Both correctly compute standing £18.63 + usage £78.40 = £97.03 pre-VAT, +5% = £101.88 with clear steps. No constraints breached. Equally useful and clear for non-technical user. | deepseek/deepseek-v4-pro: tie — Both responses correctly calculate the bill before VAT (£97.03) and with 5% VAT (£101.88), showing each step clearly. No constraints violated. Both are accurate, useful, and concise. / Both responses accurately compute the bill before VAT (£97.03) and with 5% VAT (£101.88), showing clear steps. No constraints broken. Equally useful and concise.
tieDifficult message · 8 v 8Both meet word limit and refuse clearly without false excuses. Response 1 is more concise (54 words) and firmly closes the door. Response 2 (89…
A friend has asked to borrow £2,000 and I do not want to lend it. Write what I should send. Warm, clear, no false excuses, does not leave the door open. Maximum 90 words.
Hi [Name], thanks for asking me. I’m not able to lend money, so I won’t be able to lend you the £2,000. I know that may be disappointing, and I’m sorry I can’t help in that way. I hope you understand, and I wish you the best with sorting things out.
Hi [Name], I really value our friendship and appreciate you trusting me enough to ask. However, I won’t be able to lend you the £2,000. I have a strict personal rule against lending money to friends, as I never want finances to complicate my relationships. I know this isn't the answer you were hoping for, and I hope you can understand my boundary. I care about you and am always here to support you in other ways. Let’s catch up soon. Warmly, [Your Name]
anthropic/claude-sonnet-5: tie (flipped on swap) — Both meet word limit and refuse clearly without false excuses. Response 1 is more concise (54 words) and firmly closes the door. Response 2 (89 words) is warmer but 'let's catch up soon' slightly softens closure. Both fine, but Response 1 is cleaner and more direct. / Both stay under 90 words with no false excuses and close the door. R1 is warmer, more explicit about a firm rule, and offers ongoing friendship without reopening the loan request. R2 is concise but slightly vaguer and less warm. | x-ai/grok-4.5: tie (flipped on swap) — Resp1 is warm, direct, excuse-free, firmly closed, concise. Resp2 warmer but opens door via other support/catch-up and is wordier. / Both under 90 words, warm/clear/honest/no open door. R1 more complete and supportive without excuses; R2 solid but briefer and less warm. | deepseek/deepseek-v4-pro: tie (flipped on swap) — Response 1 is warm, clear, avoids excuses entirely, and firmly closes the door. Response 2 adds a personal rule that may be a false excuse, slightly violating the 'no false excuses' instruction. Both are within word limit. / Response 1 is warmer, uses a personal rule to firmly close the door, and offers non-financial support. Response 2 is concise but less warm and may leave the door open by saying 'not able to lend money' without a permanent reason. Both under 90 words.
What they cost
- plan pro$100/mochatgpt.com/pricing8 Aug 2026verified
- plan ai plus£7.99/moone.google.com/about/plans8 Aug 2026verified
- plan ai pro£18.99/moone.google.com/about/plans8 Aug 2026verified
Speed, measured
Median response time across all 18 tasks: 6.5s (ChatGPT) vs 14.3s (Gemini).
Receipts — every citation, raw outputs
The full run — prompts, both raw outputs per task, judge reasoning, latencies, token counts and per-call cost — is downloadable as JSON. The task suite is versioned (suite-2026-08-skills/everyday_chat) and published on the methodology page.
- judge panelanthropic/claude-sonnet-5, x-ai/grok-4.5, deepseek/deepseek-v4-proour run (raw outputs)7 Aug 2026verified
- judge swap agreement0.444our run (raw outputs)7 Aug 2026verified
- judge swap kappa0.174our run (raw outputs)7 Aug 2026verified
- median latency ms a6468our run (raw outputs)7 Aug 2026verified
- median latency ms b14346our run (raw outputs)7 Aug 2026verified
- panel swap flip rate0.352our run (raw outputs)7 Aug 2026verified
- panel unanimous rate0.611our run (raw outputs)7 Aug 2026verified
- run cost a usd0.0352our run (raw outputs)7 Aug 2026verified
- run cost b usd0.3692our run (raw outputs)7 Aug 2026verified
- score a3our run (raw outputs)7 Aug 2026verified
- score b1our run (raw outputs)7 Aug 2026verified
- suite winseveryday chat: a 3/b 1/tie 14our run (raw outputs)7 Aug 2026verified
- tasks total18our run (raw outputs)7 Aug 2026verified
- ties14our run (raw outputs)7 Aug 2026verified
ChatGPT 3, Gemini 1. That margin is inside the noise of our suite, so we do not call a winner on quality. The judge could pick a winner on 4 of 18 tasks; on the other 14 it could not tell them apart.