Compare / Gemini 3.8 Flash vs Kimi K3.1 Pro

Gemini 3.8 Flash vs Kimi K3.1 Pro

Which is cheaper, which is more capable, and which does better at each kind of work, from 85 results on 6 sources that measured both models.

GoogleGemini 3.8 Flash Released 2 Sep 2026 · $1.50 in · $7.50 out
Moonshot AIKimi K3.1 Pro Released 23 Oct 2026 · $1.75 in · $10.00 out
Gemini 3.8 Flash wins 3014 too close to callKimi K3.1 Pro wins 41

Of the 85 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.

Results as of 19 November 2026 (catalogue release 2026-11-19-3608133e8331). See also Google vs Moonshot AI, every model from both labs.

Gemini 3.8 Flash or Kimi K3.1 Pro? The short answer

Pick Kimi K3.1 Pro if quality matters most: it scores higher (Spring Prompt overall 67 against 57). Pick Gemini 3.8 Flash if cost matters more: it costs 4.4 times less per product listing on CatalogBench.

  • Spring Prompt overall: Gemini 3.8 Flash 57, Kimi K3.1 Pro 67 (out of 100; how the score works).
  • Gemini 3.8 Flash is cheaper to run: $0.010 against $0.045 per product listing on CatalogBench (4.4 times less).
  • Kimi K3.1 Pro is newer: released 23 Oct 2026, against 2 Sep 2026 for Gemini 3.8 Flash.
  • On our own benchmarks, on the results both models have: Gemini 3.8 Flash does better for user surveys and speed under pressure; Kimi K3.1 Pro for decks from an analysis and marketing planning.
  • On third-party benchmarks, on the results both models have: Kimi K3.1 Pro does better for agents and tool use and reasoning and knowledge.

Head to head

57
Spring Prompt overall
67
$3.00
Price per million tokens, blendedlower is better
$3.81
1,048,576
Context window, tokens
1,048,576

Every fact side by side

Gemini 3.8 FlashKimi K3.1 Pro
Price per million tokens$1.50 in · $7.50 out$1.75 in · $10.00 out
Blended (3 in : 1 out)$3.00$3.81
Cheapest host, blended$1.50 (Google)$2.38 (Relace, FP4)
Cost per product listing (CatalogBench)$0.010$0.045
Intelligence Index40.9–
Spring Prompt overall5767
Output speed, tokens a second141–
Context window1,048,576 tokens1,048,576 tokens
Released2 Sep 202623 Oct 2026 (newer)
WeightsClosedOpen
Developer based inthe USChina

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 25 of 35 not measured
Decks from an analysisTurning a finished analysis into a deck you could present as it is 29 of 35 14 of 35
User surveysPlanning a user survey and reading its results without being misled 18 of 26 23 of 26
Marketing planningPlanning a year of ad spend without overspending 13 of 35 5 of 35
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 35 of 159 1 of 159
Professional workReal tasks from banking, consulting and law, business documents and freelance projects 11 of 51 not measured
Reasoning and knowledgeHard questions across science, maths and general knowledge 15 of 187 9 of 187
WritingWhat people prefer in blind comparisons, and judged writing quality 3 of 169 not measured
Sticking to the factsSummarising without inventing things, and factual answers 12 of 139 not measured
Speed under pressureGood decisions against a real clock (fast chess) 14 of 38 29 of 38

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.

First slide of Gemini 3.8 Flash's deck for Hearthside Coffee
Gemini 3.8 Flash · deck rating 771 · needs work
First slide of Kimi K3.1 Pro's deck for Hearthside Coffee
Kimi K3.1 Pro · deck rating 1,223 · needs work

Every slide, side by side →

From our research

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.

MeasureGemini 3.8 FlashKimi K3.1 Pro
BulletBench measured by Spring Prompt · 3 results
Cost per game · Bullet 60sUS dollars, lower is better 43 · 48 of 62 measured $0.0154 $0.0175low reasoning
Ladder Elo · Bullet 60sladder Elo, higher is better 1 · 54 of 62 measured 1,068low reasoning 147low reasoning
Games lost on time · Bullet 60s% of games, lower is better 39 · 57 of 62 measured 50.0%low reasoning 87.5%low reasoning
CatalogBench measured by Spring Prompt · 3 results
Cost per productUS dollars, lower is better 19 · 32 of 36 measured $0.0103 $0.0454
Channel compliance% of products, higher is better 28 · 1 of 36 measured 99.4% 100.0%
Channel rules broken% of products, lower is better 22 · 1 of 36 measured 0.6% 0.0%
DeckBench measured by Spring Prompt · 3 results
Cost per deckUS dollars, lower is better 15 · 32 of 36 measured $0.0613 $0.54
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. 29 · 14 of 36 measured 771 1,223
Accurate decks% of tasks, higher is better 27 · 11 of 36 measured 16.7% 66.7%
ROASBench measured by Spring Prompt · 3 results
Overall scorescore out of 100, higher is better 13 · 5 of 36 measured 51.4 55.1
Audience scorescore out of 100, higher is better 13 · 11 of 36 measured 67.1 67.6
Business scorescore out of 100, higher is better 12 · 4 of 36 measured 51.0 58.2
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. 18 · 24 of 27 measured 845 605
Cost per taskUS dollars, lower is better 14 · 25 of 27 measured $0.0943 $0.74
Analysis scorepoints of 100, higher is better 17 · 26 of 27 measured 74.7 14.5
Artificial Analysis reported by Artificial Analysis · 3 results
Humanity's Last Exam% of questions, higher is better 18 · 9 of 250 measured 47.8%high reasoning 52.5%
Terminal-Bench 4.0% of tasks, higher is better 30 · 1 of 120 measured 19.7%high reasoning 67.2%
Long-context reasoning (AA-LCR)% of questions, higher is better 7 · 3 of 241 measured 84.0%medium reasoning 86.3%
Show all 85 results (67 more)
MeasureGemini 3.8 FlashKimi K3.1 Pro
BulletBench measured by Spring Prompt · 2 results
Invalid moves · Bullet 60s% of moves, lower is better 1 · 1 of 62 measured 0.0% 0.0%low reasoning
Median move time · Bullet 60smilliseconds, lower is better 46 · 62 of 62 measured 1.8 slow reasoning 5.0 slow reasoning
CatalogBench measured by Spring Prompt · 26 results
Channel rules broken · sales brief% of products, lower is better 19 · 1 of 36 measured 0.6% 0.0%
Claims to check% of products, lower is better 1 · 1 of 34 measured 0.0% 0.0%
Claims to check · sales brief% of products, lower is better 29 · 1 of 34 measured 5.4% 0.0%
Conflicts caught% of conflicts, higher is better 1 · 1 of 36 measured 100.0% 100.0%
Content quality% of checks, higher is better 11 · 1 of 36 measured 97.7% 98.9%
Content quality · sales brief% of checks, higher is better 17 · 30 of 36 measured 97.6% 95.3%
Decision accuracy% of decisions, higher is better 21 · 1 of 36 measured 96.4% 98.9%
Failed outputs% of products, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Failed outputs · sales brief% of products, lower is better 1 · 34 of 36 measured 0.0% 3.6%
Field accuracy% of missing fields, higher is better 27 · 3 of 36 measured 92.7% 95.6%
Invented values% of filled values, lower is better 4 · 26 of 36 measured 0.3% 2.1%
Missing UK information% of products, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Missing UK information · sales brief% of products, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Not findable% of products, lower is better 30 · 21 of 36 measured 8.3% 5.4%
Not findable · sales brief% of products, lower is better 31 · 17 of 36 measured 8.9% 3.6%
Publish-ready listings% of products, higher is better 24 · 4 of 36 measured 54.8% 78.6%
Publish-ready listings · sales brief% of products, higher is better 26 · 3 of 36 measured 16.7% 75.0%
Time per productseconds, lower is better 21 · 36 of 36 measured 18.7 114
Unsupported claimsclaims per product, lower is better 25 · 1 of 36 measured 0.79 0.00
Unsupported claims% of products, lower is better 27 · 1 of 36 measured 31.6% 0.0%
Unsupported claims · sales briefclaims per product, lower is better 27 · 7 of 36 measured 3.08 0.21
Unsupported claims · sales brief% of products, lower is better 31 · 2 of 36 measured 81.6% 1.8%
Wrong attributes% of products, lower is better 14 · 18 of 36 measured 15.5% 16.1%
Wrong attributes · sales brief% of products, lower is better 2 · 6 of 36 measured 11.9% 14.3%
Wrong category or variant% of products, lower is better 30 · 7 of 36 measured 3.6% 1.8%
Wrong category or variant · sales brief% of products, lower is better 26 · 33 of 36 measured 3.0% 5.4%
DeckBench measured by Spring Prompt · 14 results
Caveat dropped% of tasks, lower is better 1 · 16 of 36 measured 0.0% 16.7%
Clean layout% of tasks, higher is better 26 · 31 of 36 measured 16.7% 0.0%
Design quality% of the maximum, higher is better 21 · 5 of 36 measured 61.7% 83.9%
Draft figure quoted% of tasks, lower is better 29 · 1 of 36 measured 50.0% 0.0%
Findings missing% of tasks, lower is better 1 · 33 of 36 measured 0.0% 33.3%
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. 29 · 10 of 36 measured 37.8% 70.8%
Layout defects% of tasks, lower is better 26 · 31 of 36 measured 83.3% 100.0%
Misleading metric used% of tasks, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Presentable decks% of tasks, higher is better 17 · 17 of 36 measured 0.0% 0.0%
Recommendation late or wrong% of tasks, lower is better 22 · 1 of 36 measured 33.3% 0.0%
Slides needing work% of tasks, lower is better 12 · 1 of 36 measured 33.3% 0.0%
Time per deckseconds, lower is better 15 · 36 of 36 measured 93.0 1,288
Unsupported claims% of tasks, lower is better 32 · 1 of 36 measured 50.0% 0.0%
Unsupported numbers% of tasks, lower is better 28 · 1 of 36 measured 16.7% 0.0%
ROASBench measured by Spring Prompt · 6 results
Consistency scorescore out of 100, higher is better 8 · 9 of 36 measured 36.4 36.3
Contribution profitsimulated US dollars, higher is better 5 · 4 of 36 measured $842,432 $858,023
Cost of a runUS dollars, lower is better 15 · 34 of 36 measured $0.16 $1.60
Months over budgetmonths of 12, lower is better 1 · 1 of 36 measured 0 0
Planning scorescore out of 100, higher is better 23 · 5 of 36 measured 54.7 55.0
Return on ad spendprofit per $1 spent, higher is better 13 · 2 of 36 measured $0.65 $0.72
SurveyBench measured by Spring Prompt · 18 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. 7 · 20 of 27 measured 60.9% 36.3%
Leading or double-barrelled questions% of tasks, lower is better 24 · 10 of 27 measured 66.7% 22.2%
Mistaken next steps% of tasks, lower is better 5 · 24 of 27 measured 11.1% 77.8%
Numbers right% of answers, higher is better 1 · 1 of 27 measured 100.0% 100.0%
Objectives unanswerable% of tasks, lower is better 1 · 1 of 27 measured 0.0% 0.0%
Output tokens per tasktokens, lower is better 5 · 17 of 17 measured 21,981 70,935
Precision ignored% of tasks, lower is better 21 · 1 of 27 measured 44.4% 0.0%
Research scorepoints of 100, higher is better 17 · 25 of 27 measured 77.6 39.2
Sound research% of tasks, higher is better 17 · 17 of 27 measured 0.0% 0.0%
Stakeholder's view presumed% of tasks, lower is better 18 · 1 of 27 measured 22.2% 0.0%
Survey plan scorepoints of 100, higher is better 21 · 4 of 27 measured 84.4 97.0
Survey structure% of tasks, lower is better 12 · 3 of 27 measured 22.2% 11.1%
Time per taskseconds, lower is better 10 · 27 of 27 measured 131 1,956
Traps handled% of traps, higher is better 21 · 14 of 27 measured 78.3% 93.3%
Traps missed% of tasks, lower is better 19 · 19 of 27 measured 88.9% 88.9%
Unsupported findings% of tasks, lower is better 16 · 16 of 27 measured 77.8% 77.8%
Wrong numbers% of tasks, lower is better 1 · 22 of 27 measured 0.0% 77.8%
Wrong recommendation% of tasks, lower is better 1 · 27 of 27 measured 0.0% 100.0%
Artificial Analysis reported by Artificial Analysis · 1 result
SciCode% of problems, higher is better 20 · 7 of 121 measured 56.6%high reasoning 60.8%

Quick answers

Is Gemini 3.8 Flash better than Kimi K3.1 Pro?

Pick Kimi K3.1 Pro if quality matters most: it scores higher (Spring Prompt overall 67 against 57). Pick Gemini 3.8 Flash if cost matters more: it costs 4.4 times less per product listing on CatalogBench.

Which is cheaper, Gemini 3.8 Flash or Kimi K3.1 Pro?

Gemini 3.8 Flash is cheaper to run: $0.010 against $0.045 per product listing on CatalogBench (4.4 times less).

Which is better for decks from an analysis, Gemini 3.8 Flash or Kimi K3.1 Pro?

Kimi K3.1 Pro ranks 14 of 35 models on sale for turning a finished analysis into a deck you could present as it is, against 29 for Gemini 3.8 Flash.

Which is better for user surveys, Gemini 3.8 Flash or Kimi K3.1 Pro?

Gemini 3.8 Flash ranks 18 of 26 models on sale for planning a user survey and reading its results without being misled, against 23 for Kimi K3.1 Pro.

Which is better for marketing planning, Gemini 3.8 Flash or Kimi K3.1 Pro?

Kimi K3.1 Pro ranks 5 of 35 models on sale for planning a year of ad spend without overspending, against 13 for Gemini 3.8 Flash.

Which is better for agents and tool use, Gemini 3.8 Flash or Kimi K3.1 Pro?

Kimi K3.1 Pro ranks 1 of 159 models on sale for multi-step tasks with tools: support desks, coding agents, function calls, against 35 for Gemini 3.8 Flash.

Which is better for reasoning and knowledge, Gemini 3.8 Flash or Kimi K3.1 Pro?

Kimi K3.1 Pro ranks 9 of 187 models on sale for hard questions across science, maths and general knowledge, against 15 for Gemini 3.8 Flash.

Run Gemini 3.8 Flash and Kimi K3.1 Pro on your own prompt.

Benchmarks aren't your task. Test both side by side on your prompt, with every answer checked against your rules and the cost of each one.