Gemini 3.1 Pro Preview
VS
Kimi K2.6
Gemini 3.1 Pro Preview vs Kimi K2.6: which wins at real work?
22 task areas · same graded test runs · rank comparison only, so 0–100 and Elo collections never mix raw scores.
Gemini 3.1 Pro Preview and Kimi K2.6 split the 22 task areas we tested 11–11. Kimi K2.6 costs 3.4× less per token ($4.07 vs $14 per 1M).
Kimi K2.6 costs 3.4× less per token ($4.07 vs $14 per 1M).
Task by task
| Task area | Gemini 3.1 Pro Preview | Kimi K2.6 | Winner |
|---|---|---|---|
| Coding |
#64
/ 115
Strong
|
#5
/ 115
Excellent
|
Kimi K2.6 |
| Training & Education |
#4
/ 110
Excellent
|
#45
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
| Product & Project Management |
#25
/ 110
Excellent
|
#53
/ 110
Strong
|
Gemini 3.1 Pro Preview |
| Frontend & Landing Pages |
#34
/ 109
Needs editing
|
#8
/ 109
Usable
|
Kimi K2.6 |
| Research & Competitive Analysis |
#42
/ 110
Strong
|
#19
/ 110
Excellent
|
Kimi K2.6 |
| Presentations & Decks |
#24
/ 110
Excellent
|
#2
/ 110
Excellent
|
Kimi K2.6 |
| Sales |
#32
/ 110
Strong
|
#15
/ 110
Strong
|
Kimi K2.6 |
| Data & Analytics |
#39
/ 110
Excellent
|
#23
/ 110
Excellent
|
Kimi K2.6 |
| AI Strategy |
#33
/ 126
Strong
|
#19
/ 126
Strong
|
Kimi K2.6 |
| Investor & Pitch |
#13
/ 66
Strong
|
#1
/ 66
Strong
|
Kimi K2.6 |
| Content & Brand |
#11
/ 124
Strong
|
#22
/ 124
Strong
|
Gemini 3.1 Pro Preview |
| RAG, Safety & Grounding |
#8
/ 113
Excellent
|
#18
/ 113
Excellent
|
Gemini 3.1 Pro Preview |
| Translation & Localization |
#14
/ 110
Excellent
|
#22
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
| Creative & Comedy | #4 / 110 | #11 / 110 | Gemini 3.1 Pro Preview |
| Landing Pages |
#14
/ 72
Strong
|
#21
/ 72
Strong
|
Gemini 3.1 Pro Preview |
| Executive Assistant |
#14
/ 112
Strong
|
#8
/ 112
Strong
|
Kimi K2.6 |
| Legal & HR |
#17
/ 110
Excellent
|
#22
/ 110
Excellent
|
Gemini 3.1 Pro Preview |
| Knowledge & Docs |
#19
/ 110
Strong
|
#22
/ 110
Strong
|
Gemini 3.1 Pro Preview |
| Customer Support |
#2
/ 113
Strong
|
#4
/ 113
Strong
|
Gemini 3.1 Pro Preview |
| Chef / Home Cooking |
#6
/ 126
Strong
|
#7
/ 126
Strong
|
Gemini 3.1 Pro Preview |
| Structured Output |
#15
/ 113
Excellent
|
#14
/ 113
Excellent
|
Kimi K2.6 |
| Summarization & Meeting Notes |
#4
/ 110
Excellent
|
#3
/ 110
Excellent
|
Kimi K2.6 |
Rank = position among every model config we tested in that task area (lower is better). Sorted by biggest gap first.
Same task, both models — judged
Both models answered the same test case; an independent judge graded each. Quotes are the judge's actual rationale.
Research & Competitive Analysis
Synthesize conflicting sources (Northwind) (Grounded Synthesis)
“The model successfully grounds all claims in the provided sources, correctly attributes them, and flags the primary conflict regarding the customer count. However, it falls short of a perfect score because it fails to explicitly weight Source C as low-reliability and does not flag the '5% guarantee' as an uncorroborated claim, instead repeating it as a reported fact.”
“The model perfectly follows all instructions. It grounds every claim in the provided sources, accurately attributes them, explicitly flags the conflict between Source A and Source C regarding customer counts, and appropriately frames Source C as an uncited and undated blog post.”
Frequently asked
Is Gemini 3.1 Pro Preview better than Kimi K2.6?
They tie across the 22 task areas we benchmarked, at 11 apiece.
Which is cheaper, Gemini 3.1 Pro Preview or Kimi K2.6?
Kimi K2.6 costs 3.4× less per token ($4.07 vs $14 per 1M).
What is Gemini 3.1 Pro Preview better at?
Gemini 3.1 Pro Preview out-ranks Kimi K2.6 at Training & Education, Product & Project Management, Content & Brand.
What is Kimi K2.6 better at?
Kimi K2.6 out-ranks Gemini 3.1 Pro Preview at Coding, Frontend & Landing Pages, Research & Competitive Analysis.
More comparisons
This page is Spring Prompt, running in public
We just did this for every model. Do it for your prompt.
The rankings above come from running real tasks through real models and scoring every output. Spring Prompt is that same engine — pointed at your prompt, your test cases, and your definition of good.
- Generate test cases from your prompt — no eval set required to start.
- Compare models side by side with quality, cost and latency in one matrix.
- Optimise the winner until the scores say it's ready to ship.
Prompt × model results
12 test cases · 3 evals