Model comparisons - tested October 2026 by Malte Pruser. How we test.
Verdict
Grok 4.6 takes this round 4 tasks to 2 with four ties, and the margins were often narrow. It won the product email by hitting the requested word count exactly, the slugify task with more meaningful assertions, the cent bug by avoiding Gemini's false Number.EPSILON claim, and the SaaS projection by keeping rounded and unrounded figures straight. Gemini 3.1 Pro won the simple explanation by making exponential versus linear explicit, and the revenue analysis by covering both readings of best and worst. The four splits - VAT, populations, water jugs and the dinner split - were genuine dead heats, and several of Grok's wins came down to small faithfulness details rather than a clear capability gap. For everyday writing and arithmetic, pick Grok 4.6; for explanatory and analytical depth, Gemini 3.1 Pro.
Verdict by task
| Task | Winner | Why |
|---|
| Writing | Split | Grok took the email by landing exactly on 120 words while Gemini ran 113; Gemini took the rewrite back by making exponential versus linear growth more explicit for a young reader. |
| Coding | Grok 4.6 | Both slugify implementations pass the required tests, but Grok's assertions are more meaningful and its mappings broader. On the cent bug the fixes are equivalent, while Gemini's claim that Number.EPSILON reliably fixes decimal rounding is wrong. |
| Research | Split | Each VAT answer has real omissions - Grok is concise with Article 59c, Gemini explains the 2020/285 SME conditions - and both population lists drift from the 2024 UN estimates they cite. |
| Data analysis | Split | Gemini covered both readings of best and worst on revenue and gave arithmetic and compound averages; Grok gave the cleaner SaaS projection while Gemini's exact table mixed rounded and unrounded figures. |
| Reasoning | Split | Dead heats on both: the same six-step jug solution and the same correct 41/73/73 dinner split with proper checks. |
Results, prompt by prompt
writing - winner: Grok 4.6
Prompt: Write a 120 word product update email to existing customers of a small accounting software company announcing that bank feeds now sync every hour instead of once a day. Plain, friendly, no exclamation marks, end with one clear call to action.
- Grok 4.6 - pass, no hallucinations, 62.3s. Body is exactly 120 words and stays faithful to the brief.
- Gemini 3.1 Pro - partial, no hallucinations, 35.4s. Friendlier tone but the body runs 113 words against the 120 requested; 120 only if you count the subject line.
writing - winner: Gemini 3.1 Pro
Prompt: Rewrite this paragraph so a 12 year old understands it, keep every fact, max 80 words: 'Compound interest is interest calculated on the initial principal and also on the accumulated interest of previous periods. Because the base grows each period, the growth is exponential rather than linear, which is why long time horizons matter far more than the interest rate alone.'
- Grok 4.6 - pass, no hallucinations, 18.5s. Meets the limit and keeps the central facts, but less explicit on exponential vs linear.
- Gemini 3.1 Pro - pass, no hallucinations, 20.5s. Makes exponential versus linear growth explicit while staying understandable.
coding - winner: Grok 4.6
Prompt: Write a Python function slugify(title: str) -> str that lowercases, replaces spaces and punctuation with single hyphens, strips accents (é -> e, ß -> ss), and returns an ASCII-only slug. Include tests for "Crème Brûlée à L'œuf" and "Hello, World!".
- Grok 4.6 - pass, no hallucinations, 32s. Handles the required examples and ß, with more meaningful assertions and broader explicit character mappings.
- Gemini 3.1 Pro - pass, no hallucinations, 31.6s. Correct on the required cases but lighter on assertions and adds unnecessary explanation.
coding - winner: Grok 4.6
Prompt: This JavaScript has a bug: users report the total is sometimes off by one cent. Find it and fix it. function total(items) { let t = 0; for (const i of items) { t += i.price * i.qty; } t += t * 0.2; return Math.round(t * 100) / 100; }
- Grok 4.6 - partial, no hallucinations, 77.7s. Main fix is equivalent to Gemini's, but does not explain the original bug.
- Gemini 3.1 Pro - partial, 1 hallucination, 24.9s. Diagnoses better, but wrongly claims adding Number.EPSILON reliably fixes decimal rounding.
research - winner: Split
Prompt: What is the current EU VAT registration threshold for small businesses selling online across EU countries in 2025? Give the figure and the rule it comes from.
- Grok 4.6 - partial, 1 hallucination, 33.5s. Concise and names Article 59c, but imprecise on the SME exemption conditions.
- Gemini 3.1 Pro - partial, 1 hallucination, 26.8s. Explains the SME exemption and cites Directive 2020/285, with its own omissions and imprecise wording.
research - winner: Split
Prompt: List the 5 largest countries by population in 2024 with their approximate population and a source you would cite.
- Grok 4.6 - partial, 1 hallucination, 25.9s. Right ranking, but the figures do not consistently match the cited 2024 UN revision.
- Gemini 3.1 Pro - partial, 1 hallucination, 12.3s. Right ranking, same problem: figures drift from the cited 2024 UN estimates.
data - winner: Gemini 3.1 Pro
Prompt: Here are monthly revenue figures in EUR for a shop: Jan 41200, Feb 38900, Mar 45100, Apr 42800, May 47600, Jun 51200. Compute month-over-month growth, identify the best and worst month, and tell me the average monthly growth rate.
- Grok 4.6 - partial, no hallucinations, 32.3s. Calculates correctly but only answers best/worst by one interpretation.
- Gemini 3.1 Pro - pass, no hallucinations, 23.3s. Distinguishes best revenue from best growth and gives both arithmetic and compound averages.
data - winner: Grok 4.6
Prompt: A SaaS company has 1,240 customers paying an average of 49 EUR per month. Monthly churn is 3.1% and they add 85 new customers each month. Project the customer count and MRR after 12 months.
- Grok 4.6 - pass, no hallucinations, 46.6s. Same headline result, shown with the recurrence worked through clearly.
- Gemini 3.1 Pro - partial, 1 hallucination, 59s. Adds an "exact" table that mixes unrounded customer counts with MRR from rounded headcounts.
reasoning - winner: Split
Prompt: I have a 3 litre jug and a 5 litre jug and unlimited water. Give the shortest sequence of steps to end with exactly 4 litres in the 5 litre jug.
- Grok 4.6 - pass, no hallucinations, 34.8s. Correct shortest six-step solution under standard jug-puzzle rules.
- Gemini 3.1 Pro - pass, no hallucinations, 23.6s. Correct shortest six-step solution under standard jug-puzzle rules.
reasoning - winner: Split
Prompt: Three colleagues need to split a 187 EUR dinner. Anna had no drinks and only ate a 41 EUR main. Ben and Clara shared a 58 EUR bottle of wine equally and each ate a 44 EUR main. Who owes what?
- Grok 4.6 - pass, no hallucinations, 8.7s. Correctly assigns 41, 73 and 73 EUR with an appropriate explanation.
- Gemini 3.1 Pro - pass, no hallucinations, 8.2s. Correctly assigns 41, 73 and 73 EUR with an appropriate explanation.
Frequently asked questions
Which is better, Grok or Gemini?
In this round Grok 4.6 edged it 4 tasks to 2 with four ties, and most margins were small. Grok is more faithful to a brief and keeps arithmetic cleaner; Gemini gives more complete explanations and broader analytical coverage.
Which is cheaper, Grok or Gemini?
Grok 4.6 by roughly 2x: about $0.13 across the ten prompts versus about $0.31 for Gemini 3.1 Pro, measured on the API. Gemini answered faster on average (about 28 seconds versus 34).
Is Grok or Gemini better for coding?
Grok 4.6, narrowly. Both slugify implementations pass the required tests, but Grok's assertions are more meaningful, and on the cent bug Gemini claims Number.EPSILON reliably fixes decimal rounding, which it does not.
Can I use Grok and Gemini in the same place?
Yes. Krater gives you Grok 4.6, Gemini 3.1 Pro and 500+ other models in one workspace, and the Model Arena sends one prompt to several models side by side, which is how this comparison was run.
How was this comparison tested?
Both models received the same ten prompts with default settings and no system prompt in October 2026, the same set used across our tested head-to-heads. We recorded time to a complete answer and the real model cost per answer, ran the generated code through its own tests, fact checked every claim, scored completion as pass, partial or fail, and then ranked the answers blind. The full procedure is on our methodology page.
Will this comparison be updated?
Yes. Flagship comparisons are rerun monthly and whenever a provider ships a new version. The tested month and model versions are shown at the top of the page.
