Compare / Gemini 3.1 Pro Preview vs Gemini 4 Argon
Gemini 3.1 Pro Preview vs Gemini 4 Argon
Which is cheaper, which is more capable, and which does better at each kind of work, from 103 results on 7 sources that measured both models.
GoogleGemini 3.1 Pro Preview
GoogleGemini 4 Argon
Gemini 3.1 Pro Preview wins 1621 too close to callGemini 4 Argon wins 66
Of the 103 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.
Results as of 14 October 2026 (catalogue release 2026-10-14-7d7cf32c3f6d).
Gemini 3.1 Pro Preview or Gemini 4 Argon? The short answer
Pick Gemini 4 Argon: it scores higher (Spring Prompt overall 75 against 47) and costs less.
- Spring Prompt overall: Gemini 3.1 Pro Preview 47, Gemini 4 Argon 75 (out of 100; how the score works).
- Gemini 4 Argon is cheaper to run: $0.011 against $0.035 per product listing on CatalogBench (3.2 times less).
- Gemini 4 Argon scores higher on the Artificial Analysis Intelligence Index: 52.6 against 29.7.
- Gemini 4 Argon is newer: released 13 Oct 2026, against 19 Feb 2026 for Gemini 3.1 Pro Preview.
- On our own benchmarks, on the results both models have: Gemini 4 Argon does better for product listings, decks from an analysis, user surveys and marketing planning.
- On third-party benchmarks, on the results both models have: Gemini 4 Argon does better for agents and tool use, reasoning and knowledge, writing and sticking to the facts.
Head to head
47
Spring Prompt overall
75
29.7
Intelligence Index
52.6
$4.50
Price per million tokens, blendedlower is better
$4.00
1,048,576
Context window, tokens
1,048,576
Every fact side by side
| Gemini 3.1 Pro Preview | Gemini 4 Argon | |
|---|---|---|
| Price per million tokens | $2.00 in · $12.00 out | $2.00 in · $10.00 out |
| Blended (3 in : 1 out) | $4.50 | $4.00 |
| Cheapest host, blended | $4.50 (Google) | – |
| Cost per product listing (CatalogBench) | $0.035 | $0.011 |
| Cached input, per million | $0.20 cached input | – |
| Intelligence Index | 29.7 | 52.6 |
| Spring Prompt overall | 47 | 75 |
| Output speed, tokens a second | 135 | – |
| Context window | 1,048,576 tokens | 1,048,576 tokens |
| Released | 19 Feb 2026 | 13 Oct 2026 (newer) |
| Weights | Closed | Closed |
| Developer based in | the US | the US |
Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.
Which is better for what
Each model's rank among the models on sale today, for each kind of work, from our own benchmarks and licensed sources. Further right is better; the better rank is in green.
Product listingsTurning a sparse product feed and photos into listings that can go live
10 of 22
7 of 22
Decks from an analysisTurning a finished analysis into a deck you could present as it is
16 of 21
13 of 21
User surveysPlanning a user survey and reading its results without being misled
7 of 10
5 of 10
Marketing planningPlanning a year of ad spend without overspending
10 of 21
8 of 21
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls
62 of 166
3 of 166
Professional workReal tasks from banking, consulting and law, business documents and freelance projects
36 of 51
not measured
Reasoning and knowledgeHard questions across science, maths and general knowledge
26 of 200
3 of 200
WritingWhat people prefer in blind comparisons, and judged writing quality
9 of 189
1 of 189
Sticking to the factsSummarising without inventing things, and factual answers
35 of 153
1 of 153
Speed under pressureGood decisions against a real clock (fast chess)
5 of 25
20 of 25
Their slides, side by side
The same brief for both: Manchester store investment review for Hearthside Coffee, an invented company. First slide of each deck from DeckBench.
From our research
- Gemini 4 Argon review: Opus-level product data at a quarter of the cost, until you ask it to sell 14 Oct 2026
- BulletBench: we made 23 AI models play speed chess, and the clock was the judge 3 Jul 2026
- Claude vs Gemini vs GPT in a 12-Month Marketing Simulation 25 Mar 2026
- The AI Marketing Benchmark That Punishes Plausible-Sounding Strategy 20 Mar 2026
Every shared result
Each model's best result on every measure both have, grouped by benchmark, with the setting that produced it. Under each measure, every model that benchmark has measured is a grey tick, best to the right, so you can see where the two sit in the field. The headline measures of each benchmark come first; the rest are under "Show all". Green marks the better value; where a source reports ranges that overlap, neither is marked.
| Measure | Gemini 3.1 Pro Preview | Gemini 4 Argon |
|---|---|---|
| BulletBench measured by Spring Prompt · 3 results | ||
| Cost per game · Blitz 3+2US dollars, lower is better 12 · 10 of 15 measured | $0.21 | $0.12 |
| Cost per game · Bullet 60sUS dollars, lower is better 27 · 10 of 30 measured | $0.0446low reasoning | $0.0036 |
| Ladder Elo · Bullet 60sladder Elo, higher is better 11 · 26 of 30 measured | 529low reasoning | 147 |
| CatalogBench measured by Spring Prompt · 3 results | ||
| Cost per productUS dollars, lower is better 17 · 12 of 22 measured | $0.0355 | $0.0109 |
| Reliably publish-ready% of products, higher is better 10 · 5 of 22 measured | 44.6% | 60.7% |
| Reliably publish-ready · sales brief% of products, higher is better 10 · 9 of 22 measured | 16.1% | 17.9% |
| DeckBench measured by Spring Prompt · 3 results | ||
| Cost per deckUS dollars, lower is better 11 · 10 of 21 measured | $0.14 | $0.11 |
| Deck ratingrating, higher is betterThe rating weighs each win by how strong the other deck was; the win rate counts wins against whichever decks a model happened to meet, so the two can disagree. 16 · 13 of 21 measured | 763 | 998 |
| Accurate decks% of tasks, higher is better 9 · 13 of 21 measured | 50.0% | 33.3% |
| ROASBench measured by Spring Prompt · 3 results | ||
| Overall scorescore out of 100, higher is better 10 · 8 of 21 measured | 43.5 | 47.0 |
| Audience scorescore out of 100, higher is better 6 · 8 of 21 measured | 67.5 | 65.5 |
| Business scorescore out of 100, higher is better 10 · 8 of 21 measured | 35.9 | 43.3 |
| SurveyBench measured by Spring Prompt · 3 results | ||
| Analysis ratingrating, higher is betterThe rating weighs each win by how strong the other analysis was; the win rate counts wins against whichever analyses a model happened to meet, so the two can disagree. 7 · 5 of 10 measured | 776 | 964 |
| Cost per taskUS dollars, lower is better 8 · 6 of 10 measured | $0.33 | $0.13 |
| Analysis scorepoints of 100, higher is better 6 · 5 of 10 measured | 66.4 | 79.6 |
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results | ||
| Overall · TextArena rating, higher is better 15 · 1 of 387 measured | 1,487 | 1,525high reasoning |
| Overall · Text factualityArena rating, higher is better 21 · 1 of 163 measured | 1,472 | 1,511high reasoning |
| Business, management and finance · TextArena rating, higher is better 23 · 1 of 380 measured | 1,477 | 1,522high reasoning |
| Artificial Analysis reported by Artificial Analysis · 3 results | ||
| Artificial Analysis Intelligence Indexindex score, higher is better 53 · 5 of 244 measured | 29.7 | 52.6high reasoning |
| Humanity's Last Exam% of questions, higher is better 17 · 3 of 243 measured | 47.0% | 57.1%high reasoning |
| Terminal-Bench 4.0% of tasks, higher is better 46 · 4 of 113 measured | 4.0% | 57.1%high reasoning |
Show all 103 results (82 more)
| Measure | Gemini 3.1 Pro Preview | Gemini 4 Argon |
|---|---|---|
| BulletBench measured by Spring Prompt · 7 results | ||
| Games lost on time · Blitz 3+2% of games, lower is better 12 · 1 of 15 measured | 50.0% | 0.0% |
| Games lost on time · Bullet 60s% of games, lower is better 23 · 28 of 30 measured | 58.3%low reasoning | 87.5% |
| Invalid moves · Blitz 3+2% of moves, lower is better 1 · 1 of 15 measured | 0.0% | 0.0% |
| Invalid moves · Bullet 60s% of moves, lower is better 1 · 1 of 30 measured | 0.0%low reasoning | 0.0% |
| Ladder Elo · Blitz 3+2ladder Elo, higher is better 6 · 1 of 15 measured | 683 | 1,147 |
| Median move time · Blitz 3+2milliseconds, lower is better 11 · 5 of 15 measured | 6.2 s | 1.9 s |
| Median move time · Bullet 60smilliseconds, lower is better 29 · 28 of 30 measured | 3.1 slow reasoning | 2.9 s |
| CatalogBench measured by Spring Prompt · 28 results | ||
| Channel compliance% of products, higher is better 15 · 1 of 22 measured | 99.4% | 100.0% |
| Channel rules broken% of products, lower is better 12 · 1 of 22 measured | 0.6% | 0.0% |
| Channel rules broken · sales brief% of products, lower is better 1 · 1 of 22 measured | 0.0% | 0.0% |
| Claims to check% of products, lower is better 18 · 1 of 22 measured | 1.2% | 0.0% |
| Claims to check · sales brief% of products, lower is better 19 · 1 of 22 measured | 10.1% | 0.0% |
| Conflicts caught% of conflicts, higher is better 1 · 1 of 22 measured | 100.0% | 100.0% |
| Content quality% of checks, higher is better 12 · 4 of 22 measured | 96.9% | 98.2% |
| Content quality · sales brief% of checks, higher is better 12 · 4 of 22 measured | 97.2% | 98.4% |
| Decision accuracy% of decisions, higher is better 11 · 5 of 22 measured | 96.6% | 97.8% |
| Failed outputs% of products, lower is better 1 · 1 of 22 measured | 0.0% | 0.0% |
| Failed outputs · sales brief% of products, lower is better 1 · 1 of 22 measured | 0.0% | 0.0% |
| Field accuracy% of missing fields, higher is better 2 · 3 of 22 measured | 95.4% | 94.7% |
| Invented values% of filled values, lower is better 12 · 14 of 22 measured | 1.5% | 1.7% |
| Missing UK information% of products, lower is better 1 · 1 of 22 measured | 0.0% | 0.0% |
| Missing UK information · sales brief% of products, lower is better 1 · 1 of 22 measured | 0.0% | 0.0% |
| Not findable% of products, lower is better 15 · 3 of 22 measured | 7.7% | 2.4% |
| Not findable · sales brief% of products, lower is better 15 · 2 of 22 measured | 6.0% | 1.8% |
| Publish-ready listings% of products, higher is better 12 · 5 of 22 measured | 58.9% | 70.8% |
| Publish-ready listings · sales brief% of products, higher is better 11 · 8 of 22 measured | 30.4% | 35.1% |
| Time per productseconds, lower is better 17 · 3 of 22 measured | 22.2 | 7.30 |
| Unsupported claimsclaims per product, lower is better 11 · 6 of 22 measured | 0.66 | 0.42 |
| Unsupported claims% of products, lower is better 12 · 9 of 22 measured | 26.8% | 16.1% |
| Unsupported claims · sales briefclaims per product, lower is better 13 · 12 of 22 measured | 2.13 | 1.98 |
| Unsupported claims · sales brief% of products, lower is better 13 · 12 of 22 measured | 67.3% | 63.1% |
| Wrong attributes% of products, lower is better 5 · 6 of 22 measured | 14.9% | 15.5% |
| Wrong attributes · sales brief% of products, lower is better 10 · 7 of 22 measured | 16.1% | 15.5% |
| Wrong category or variant% of products, lower is better 4 · 4 of 22 measured | 1.8% | 1.8% |
| Wrong category or variant · sales brief% of products, lower is better 4 · 4 of 22 measured | 1.8% | 1.8% |
| DeckBench measured by Spring Prompt · 14 results | ||
| Caveat dropped% of tasks, lower is better 12 · 1 of 21 measured | 16.7% | 0.0% |
| Clean layout% of tasks, higher is better 11 · 13 of 21 measured | 50.0% | 33.3% |
| Design quality% of the maximum, higher is better 17 · 16 of 21 measured | 51.4% | 55.2% |
| Draft figure quoted% of tasks, lower is better 1 · 17 of 21 measured | 0.0% | 50.0% |
| Findings missing% of tasks, lower is better 1 · 1 of 21 measured | 0.0% | 0.0% |
| Head-to-head win rate% of comparisons, higher is betterThe rating weighs each win by how strong the other deck was; the win rate counts wins against whichever decks a model happened to meet, so the two can disagree. 16 · 13 of 21 measured | 34.1% | 57.0% |
| Layout defects% of tasks, lower is better 11 · 13 of 21 measured | 50.0% | 66.7% |
| Misleading metric used% of tasks, lower is better 1 · 1 of 21 measured | 0.0% | 0.0% |
| Presentable decks% of tasks, higher is better 10 · 12 of 21 measured | 16.7% | 0.0% |
| Recommendation late or wrong% of tasks, lower is better 15 · 1 of 21 measured | 50.0% | 0.0% |
| Slides needing work% of tasks, lower is better 19 · 15 of 21 measured | 83.3% | 66.7% |
| Time per deckseconds, lower is better 7 · 6 of 21 measured | 78.0 | 67.5 |
| Unsupported claims% of tasks, lower is better 1 · 11 of 21 measured | 0.0% | 16.7% |
| Unsupported numbers% of tasks, lower is better 1 · 1 of 21 measured | 0.0% | 0.0% |
| ROASBench measured by Spring Prompt · 6 results | ||
| Consistency scorescore out of 100, higher is better 6 · 5 of 21 measured | 33.9 | 34.7 |
| Contribution profitsimulated US dollars, higher is better 11 · 5 of 21 measured | $507,736 | $673,561 |
| Cost of a runUS dollars, lower is better 15 · 11 of 21 measured | $0.52 | $0.24 |
| Months over budgetmonths of 12, lower is better 1 · 1 of 21 measured | 0 | 0 |
| Planning scorescore out of 100, higher is better 8 · 1 of 21 measured | 54.9 | 55.4 |
| Return on ad spendprofit per $1 spent, higher is better 11 · 8 of 21 measured | $0.42 | $0.55 |
| SurveyBench measured by Spring Prompt · 17 results | ||
| Head-to-head win rate% of comparisons, higher is betterThe rating weighs each win by how strong the other analysis was; the win rate counts wins against whichever analyses a model happened to meet, so the two can disagree. 6 · 5 of 10 measured | 33.8% | 42.2% |
| Leading or double-barrelled questions% of tasks, lower is better 5 · 8 of 10 measured | 44.4% | 66.7% |
| Mistaken next steps% of tasks, lower is better 5 · 5 of 10 measured | 33.3% | 33.3% |
| Numbers right% of answers, higher is better 1 · 7 of 10 measured | 100.0% | 97.1% |
| Objectives unanswerable% of tasks, lower is better 1 · 1 of 10 measured | 0.0% | 0.0% |
| Precision ignored% of tasks, lower is better 6 · 4 of 10 measured | 55.6% | 11.1% |
| Research scorepoints of 100, higher is better 6 · 5 of 10 measured | 71.3 | 83.3 |
| Sound research% of tasks, higher is better 5 · 5 of 10 measured | 0.0% | 0.0% |
| Stakeholder's view presumed% of tasks, lower is better 6 · 4 of 10 measured | 22.2% | 11.1% |
| Survey plan scorepoints of 100, higher is better 8 · 4 of 10 measured | 82.9 | 91.8 |
| Survey structure% of tasks, lower is better 7 · 1 of 10 measured | 22.2% | 0.0% |
| Time per taskseconds, lower is better 7 · 3 of 10 measured | 170 | 74.4 |
| Traps handled% of traps, higher is better 7 · 5 of 10 measured | 71.0% | 84.1% |
| Traps missed% of tasks, lower is better 6 · 5 of 10 measured | 88.9% | 77.8% |
| Unsupported findings% of tasks, lower is better 5 · 7 of 10 measured | 77.8% | 88.9% |
| Wrong numbers% of tasks, lower is better 1 · 7 of 10 measured | 0.0% | 33.3% |
| Wrong recommendation% of tasks, lower is better 7 · 3 of 10 measured | 44.4% | 11.1% |
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 8 results | ||
| Confirmed task success · AgentIPS effect estimate, higher is better 42 · 3 of 48 measured | -0.10arena agent | 0.15high reasoning, arena agent |
| Creative writing · TextArena rating, higher is better 10 · 1 of 385 measured | 1,480 | 1,519high reasoning |
| Expert prompts · TextArena rating, higher is better 26 · 7 of 338 measured | 1,509 | 1,539high reasoning |
| Instruction following · TextArena rating, higher is better 17 · 1 of 387 measured | 1,480 | 1,528high reasoning |
| Praise over complaint · AgentIPS effect estimate, higher is better 29 · 4 of 48 measured | -0.01arena agent | 0.28high reasoning, arena agent |
| Steerability · AgentIPS effect estimate, higher is better 37 · 1 of 48 measured | -0.05arena agent | 0.13high reasoning, arena agent |
| Tool grounding · AgentIPS effect estimate, higher is better 35 · 29 of 48 measured | -0.00arena agent | 0.00high reasoning, arena agent |
| Writing, literature and language · TextArena rating, higher is better 11 · 1 of 386 measured | 1,480 | 1,522high reasoning |
| Artificial Analysis reported by Artificial Analysis · 2 results | ||
| Long-context reasoning (AA-LCR)% of questions, higher is better 24 · 44 of 234 measured | 82.0% | 79.7%high reasoning |
| SciCode% of problems, higher is better 13 · 3 of 114 measured | 58.7% | 61.8%high reasoning |
Compare them with other models
- Gemini 3.1 Pro Preview vs Claude Opus 5.5Most intelligent at $3–10 per million tokens
- Gemini 3.1 Pro Preview vs Inkling SmallFastest with similar intelligence
- Gemini 4 Argon vs Claude Opus 5.5Most intelligent at $3–10 per million tokens
- Gemini 4 Argon vs Claude Sonnet 5.5Cheapest with similar intelligence
- Gemini 4 Argon vs GPT-6 AstraJust above overall
- Gemini 4 Argon vs Muse WonderJust below overall
Quick answers
Is Gemini 3.1 Pro Preview better than Gemini 4 Argon?
Pick Gemini 4 Argon: it scores higher (Spring Prompt overall 75 against 47) and costs less.
Which is cheaper, Gemini 3.1 Pro Preview or Gemini 4 Argon?
Gemini 4 Argon is cheaper to run: $0.011 against $0.035 per product listing on CatalogBench (3.2 times less).
Which is better for product listings, Gemini 3.1 Pro Preview or Gemini 4 Argon?
Gemini 4 Argon ranks 7 of 22 models on sale for turning a sparse product feed and photos into listings that can go live, against 10 for Gemini 3.1 Pro Preview.
Which is better for decks from an analysis, Gemini 3.1 Pro Preview or Gemini 4 Argon?
Gemini 4 Argon ranks 13 of 21 models on sale for turning a finished analysis into a deck you could present as it is, against 16 for Gemini 3.1 Pro Preview.
Which is better for user surveys, Gemini 3.1 Pro Preview or Gemini 4 Argon?
Gemini 4 Argon ranks 5 of 10 models on sale for planning a user survey and reading its results without being misled, against 7 for Gemini 3.1 Pro Preview.
Which is better for marketing planning, Gemini 3.1 Pro Preview or Gemini 4 Argon?
Gemini 4 Argon ranks 8 of 21 models on sale for planning a year of ad spend without overspending, against 10 for Gemini 3.1 Pro Preview.
Which is better for agents and tool use, Gemini 3.1 Pro Preview or Gemini 4 Argon?
Gemini 4 Argon ranks 3 of 166 models on sale for multi-step tasks with tools: support desks, coding agents, function calls, against 62 for Gemini 3.1 Pro Preview.
Run Gemini 3.1 Pro Preview and Gemini 4 Argon on your own prompt
Benchmarks aren't your data. Try both side by side in the playground, with the cost of every answer.