Head-to-head comparison

Kimi K2.6 vs. Mistral Large 3

Compare published benchmark results from matching versions and cohorts, API pricing, and technical specifications. Any documented test setup differences remain visible.

shared result series: 40
distinct benchmarks: 12
latest retrieval: 2026-09-29

First modelSecond model

Specifications

General specifications

Release date, availability, context window, architecture, and published parameter counts at a glance. Undisclosed details remain marked as unavailable.

Specifications for Kimi K2.6 and Mistral Large 3
Attribute	Kimi K2.6	Mistral Large 3
Provider	Moonshot AI	Mistral AI
Released	Apr 20, 2026	Dec 2, 2025
Availability	Active	Active
Model type	Open Weights	Open Weights
Parameters	1T, 32B active	675B, 41B active
Architecture	Mixture of Experts	Mixture of Experts
Context window	262,144 tokens	262,144 tokens
Knowledge cutoff	Not published	Not published
Technical sources	Model, Context	Model, Context

Pricing

API pricing compared

The table compares official standard API rates per 1M tokens, including published cache pricing. Batch, fast, priority, regional surcharges, tool calls, and cloud platform pricing are excluded.

API pricing for Kimi K2.6 and Mistral Large 3
Price type	Kimi K2.6	Mistral Large 3
API input	$0.95	$0.5
API output	$4	$1.5
Cache read	$0.16	Not listed
Cache write (5 min.)	Not listed	Not listed
Cache write (1 hr.)	Not listed	Not listed
Price verified	09/22/2026 Source	09/22/2026 Source

Kimi K2.6Mistral Large 3

API inputUSD per 1M tokens, base tier

Kimi K2.6$0.95

Mistral Large 3$0.5

API outputUSD per 1M tokens, base tier

Kimi K2.6$4

Mistral Large 3$1.5

Key to the marks:better valuebetter, but not conclusivetie

Pricing by provider

Direct developer pricing and provider endpoint tariffs routed through OpenRouter are listed separately. Prices are in USD per 1M tokens. Context-dependent direct tiers plus regional, Flex, Priority, and other OpenRouter tariffs remain individually identifiable.

Provider pricing for Kimi K2.6 and Mistral Large 3
Model	Provider	Purchase route and tariff	Input per 1M tokens	Output per 1M tokens	Source
Kimi K2.6	Moonshot AI	Direct from developerStandard API	$0.95	$4	Source09/22/2026
	AtlasCloud	Via OpenRouteratlas-cloud/int4	$0.95	$4	Source10/02/2026
	Baidu	Via OpenRouterbaidu/fp4	Lowest listed price: $0.4342	Lowest listed price: $1.828	Source10/02/2026
	Chutes	Via OpenRouterchutes/int4	$0.5	$2.85	Source10/02/2026
	Cloudflare	Via OpenRoutercloudflare	$0.95	$4	Source10/02/2026
	CoreWeave	Via OpenRoutercoreweave/fp4	$0.65	$3.41	Source10/02/2026
	Crusoe	Via OpenRoutercrusoe/bf16	$0.7	$3.5	Source10/02/2026
	Decart	Via OpenRouterdecart/fp4	$0.5865	$2.4696	Source10/02/2026
	DeepInfra	Via OpenRouterdeepinfra/fp4	$0.75	$3.5	Source10/02/2026
	DigitalOcean	Via OpenRouterdigitalocean	$0.57	$2.4	Source10/02/2026
	GMICloud	Via OpenRoutergmicloud/fp8	$0.855	$3.6	Source10/02/2026
	Inceptron	Via OpenRouterinceptron/int4	$0.4344	$2.45	Source10/02/2026
	Moonshot AI	Via OpenRoutermoonshotai/int4	$0.95	$4	Source10/02/2026
	Novita	Via OpenRouternovita	$0.8	$3.4	Source10/02/2026
	Parasail	Via OpenRouterparasail/int4	$0.75	$3.5	Source10/02/2026
	Phala	Via OpenRouterphala	$1.09	$4.6	Source10/02/2026
	SiliconFlow	Via OpenRoutersiliconflow/fp8	$0.77	$3.4	Source10/02/2026
	StreamLake	Via OpenRouterstreamlake/fp8	$0.5985	$2.52	Source10/02/2026
Mistral Large 3	Mistral AI	Direct from developerStandard API	Lowest listed price: $0.5	Lowest listed price: $1.5	Source09/22/2026
	Mistral	Via OpenRoutermistral/eu	$0.55	$1.65	Source10/02/2026
	Mistral	Via OpenRoutermistral/zdr	Lowest listed price: $0.5	Lowest listed price: $1.5	Source10/02/2026

Performance

Shared benchmarks

Only published results with the same benchmark version, task, metric, and comparison cohort are paired. Different reasoning levels, agents, harnesses, or output limits appear directly in the table.

Capability profile from matched benchmarks

Each axis averages directly matched benchmark families. A value of 100 means the stronger value within this pair, not a universal quality score.

Kimi K2.6

Mistral Large 3

Coding: 100 / 71.7 (3 benchmark families)

Reasoning: 100 / 86.4 (2 benchmark families)

Mathematics: 100 / 94.9 (1 benchmark family)

Knowledge: 100 / 93.5 (2 benchmark families)

Finance: 100 / 91.3 (4 benchmark families)

Legal: 100 / 95.2 (2 benchmark families)

Agentic tasks: 100 / 15.7 (1 benchmark family)

Multimodal: 100 / 76.7 (1 benchmark family)

40 matched rows from 12 benchmark families were considered. The radar shows 8 of 11 comparable categories.

Kimi K2.6Mistral Large 3

CorpFin v2Overall. Higher is better. No documented setup difference

Kimi K2.666.744 %

Mistral Large 361.033 %

GPQA DiamondOverall. Higher is better. No documented setup difference

Kimi K2.689.142 %

Mistral Large 368.434 %

LegalBenchOverall. Higher is better. No documented setup difference

Kimi K2.684.738 %

Mistral Large 379.138 %

LiveCodeBench v6Overall. Higher is better. No documented setup difference

Kimi K2.686.771 %

Mistral Large 355.337 %

LMArena Text, Style ControlOverall. Higher is better. No documented setup difference

Kimi K2.61,460.921

Mistral Large 31,413.254

MMLU-ProOverall. Higher is better. No documented setup difference

Kimi K2.687.572 %

Mistral Large 379.823 %

MMMU-ProOverall. Higher is better. No documented setup difference

Kimi K2.686.301 %

Mistral Large 366.185 %

MortgageTaxOverall. Higher is better. No documented setup difference

Kimi K2.665.818 %

Mistral Large 352.106 %

SAGEOverall. Higher is better. No documented setup difference

Kimi K2.650.224 %

Mistral Large 324.595 %

SWE-bench VerifiedOverall. Higher is better. No documented setup difference

Kimi K2.676.2 %

Mistral Large 341.4 %

TaxEval v2Overall. Higher is better. No documented setup difference

Kimi K2.674.652 %

Mistral Large 373.058 %

Terminal-Bench 2.0Overall. Higher is better. No documented setup difference

Kimi K2.657.303 %

Mistral Large 38.989 %

Key to the marks:better valuebetter, but not conclusivetie

Our model tests

Our tests

Compare how the models respond to the same prompt. Each test shows the first attempt, with no subsequent fixes to the generated code. These results do not contribute to an overall score.

Pelican on a bicycle

A community classic for free-form SVG drawing.

Kimi K2.6

API access was unavailable for this run. The model was not evaluated.

Reasoning (requested): high

First attempt

Run date: 7 Sept 2026
API provider: openrouter
Requested API model: moonshotai/kimi-k2.6
Reasoning setting: high
Pinned endpoint: Moonshot AI | moonshotai/kimi-k2.6-20260420
Temperature: Provider default
Token limit including reasoning: 8,192 tokens
Test protocol: community-visual-tests-v1
Technical checks: Not run

Not yet visually reviewed. A successful recording does not confirm correct physics or full compliance with the prompt.

Mistral Large 3

API access was unavailable for this run. The model was not evaluated.

Reasoning (model default): no reasoning mode Source

First attempt

Run date: 7 Sept 2026
API provider: openrouter
Requested API model: mistralai/mistral-large-2512
Reasoning setting: not specified
Pinned endpoint: Mistral | mistralai/mistral-large-2512
Temperature: Provider default
Token limit including reasoning: 8,192 tokens
Test protocol: community-visual-tests-v1
Technical checks: Not run

Not yet visually reviewed. A successful recording does not confirm correct physics or full compliance with the prompt.

Prompt and test conditions

Generate an SVG of a pelican riding a bicycle

Token limit including reasoning: 8,192 tokens

Task origin (Simon Willison)

How to read this comparison

Winning one benchmark is not an overall verdict

Your workload, budget, and required context length matter most. A coding benchmark says little about visual understanding or agent performance.

The standard price table uses direct developer rates. The provider comparison labels OpenRouter endpoints, regions, and special tariffs separately.

Every result links to its measurement source and retrieval date. Labels distinguish vendor reports, official benchmark leaderboards, and independent evaluations. Disputed or archived results are excluded.

More matchups

Compare Kimi K2.6 and Mistral Large 3 with other leading models using the same data and benchmark logic.

View the complete LLM comparison

Benchmark scores for Kimi K2.6 and Mistral Large 3
Benchmark and task	Kimi K2.6	Mistral Large 3	Source
CorpFin v2Overall Test details No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is better	Winner66.744 %	61.033 %	Vals AI Source details Data as of: 08/12/2026
GPQA DiamondOverall Test details No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better	Winner89.142 %	68.434 %	Vals AI Source details Data as of: 09/01/2026
LegalBenchOverall Test details No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better	Winner84.738 %	79.138 %	Vals AI Source details Data as of: 09/27/2026
LiveCodeBench v6Overall Test details No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better	Winner86.771 %	55.337 %	Vals AI Source details Data as of: 09/01/2026
LMArena Text, Style ControlOverall Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 39,971 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 77,456 anonymous pairwise votes.	Winner1,460.921 95% confidence interval: ±4.435The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,413.254 95% confidence interval: ±2.912The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
MMLU-ProOverall Test details No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better	Winner87.572 %	79.823 %	Vals AI Source details Data as of: 09/01/2026
MMMU-ProOverall Test details No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better	Winner86.301 %	66.185 %	Vals AI Source details Data as of: 09/01/2026
MortgageTaxOverall Test details No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better	Winner65.818 %	52.106 %	Vals AI Source details Data as of: 09/01/2026
SAGEOverall Test details No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better	Winner50.224 %	24.595 %	Vals AI Source details Data as of: 09/26/2026
SWE-bench VerifiedOverall Test details No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better	Winner76.2 %	41.4 %	Vals AI Source details Data as of: 09/01/2026
TaxEval v2Overall Test details No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is better	Winner74.652 %	73.058 %	Vals AI Source details Data as of: 09/01/2026
Terminal-Bench 2.0Overall Test details No documented difference in the test setupVersion: 2.0Metric: accuracyScoring: Higher is better	Winner57.303 %	8.989 %	Vals AI Source details Data as of: 06/04/2026
LMArena Text, Style ControlBusiness, management and finance Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 7,948 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 14,597 anonymous pairwise votes.	Winner1,454.77 95% confidence interval: ±7.55The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,409.446 95% confidence interval: ±5.509The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlChinese Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 2,343 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 5,845 anonymous pairwise votes.	Winner1,517.112 95% confidence interval: ±12.799The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,434.058 95% confidence interval: ±8.22The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlCoding Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 11,092 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 19,574 anonymous pairwise votes.	Winner1,515.582 95% confidence interval: ±6.896The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,469.182 95% confidence interval: ±4.902The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlCreative writing Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 6,305 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 13,448 anonymous pairwise votes.	Winner1,432.89 95% confidence interval: ±8.246The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,372.648 95% confidence interval: ±5.824The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlEnglish Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 18,087 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 32,060 anonymous pairwise votes.	Winner1,469.203 95% confidence interval: ±5.649The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,426.615 95% confidence interval: ±4.05The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlEntertainment, sports and media Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 8,265 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 16,716 anonymous pairwise votes.	Winner1,428.923 95% confidence interval: ±7.544The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,371.316 95% confidence interval: ±5.299The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlExcluding ties Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 29,598 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 56,501 anonymous pairwise votes.	Winner1,464.783 95% confidence interval: ±5.815The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,400.18 95% confidence interval: ±4.058The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlExpert prompts Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 3,841 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 7,295 anonymous pairwise votes.	Winner1,502.88 95% confidence interval: ±9.953The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,427.543 95% confidence interval: ±7.482The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlFrench Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 1,575 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 2,181 anonymous pairwise votes.	Winner1,477.486 95% confidence interval: ±16.574The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,432.389 95% confidence interval: ±14.389The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlGerman Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 746 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 1,257 anonymous pairwise votes.	Winner1,457.617 95% confidence interval: ±21.288The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,409.467 95% confidence interval: ±17.833The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlHard prompts Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 26,094 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 47,316 anonymous pairwise votes.	Winner1,485.462 95% confidence interval: ±5.188The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,431.274 95% confidence interval: ±3.63The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlHard prompts, English Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 12,459 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 19,638 anonymous pairwise votes.	Winner1,488.176 95% confidence interval: ±6.532The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,439.274 95% confidence interval: ±4.91The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlInstruction following Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 13,226 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 24,861 anonymous pairwise votes.	Winner1,455.711 95% confidence interval: ±6.32The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,400.115 95% confidence interval: ±4.491The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlJapanese Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 498 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 830 anonymous pairwise votes.	Winner1,441.882 95% confidence interval: ±28.159The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,382.659 95% confidence interval: ±22.106The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlKorean Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 730 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 1,318 anonymous pairwise votes.	Winner1,423.745 95% confidence interval: ±22.737The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,356.821 95% confidence interval: ±17.794The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlLegal and government Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 3,028 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 5,879 anonymous pairwise votes.	Winner1,464.097 95% confidence interval: ±11.113The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,421.357 95% confidence interval: ±8.211The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlLife, physical and social science Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 6,726 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 12,690 anonymous pairwise votes.	Winner1,478.057 95% confidence interval: ±7.876The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,428.685 95% confidence interval: ±5.734The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlLonger queries Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 16,975 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 29,146 anonymous pairwise votes.	Winner1,477.914 95% confidence interval: ±6.084The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,417.29 95% confidence interval: ±4.417The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlMath Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 2,076 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 4,411 anonymous pairwise votes.	Winner1,479.908 95% confidence interval: ±13.073The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,401.493 95% confidence interval: ±9.177The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlMathematical professions Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 2,216 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 4,047 anonymous pairwise votes.	Winner1,485.541 95% confidence interval: ±12.814The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,413.935 95% confidence interval: ±9.781The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlMedicine and healthcare Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 2,934 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 5,443 anonymous pairwise votes.	Winner1,473.129 95% confidence interval: ±11.438The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,430.076 95% confidence interval: ±8.59The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlMulti-turn Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 6,946 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 13,164 anonymous pairwise votes.	Winner1,460.669 95% confidence interval: ±7.98The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,419.382 95% confidence interval: ±5.74The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlNon-English Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 21,884 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 45,384 anonymous pairwise votes.	Winner1,448.57 95% confidence interval: ±5.333The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,397.089 95% confidence interval: ±3.587The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlPolish Test details No documented difference in the test setupNo clear leadVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 761 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 1,754 anonymous pairwise votes.	No clear lead1,443.087 95% confidence interval: ±20.906The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,414.116 95% confidence interval: ±14.298The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlRussian Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 4,254 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 9,292 anonymous pairwise votes.	Winner1,453.596 95% confidence interval: ±9.384The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,408.043 95% confidence interval: ±6.505The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlSoftware and IT services Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 16,059 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 29,068 anonymous pairwise votes.	Winner1,502.848 95% confidence interval: ±5.972The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,455.584 95% confidence interval: ±4.237The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlSpanish Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 1,379 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 2,328 anonymous pairwise votes.	Winner1,458.419 95% confidence interval: ±16.701The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,405.285 95% confidence interval: ±13.338The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026
LMArena Text, Style ControlWriting, literature and language Test details No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterKimi K2.6: Snapshot: kimi-k2.6, Harness: Arena text, style control Mistral Large 3: Snapshot: mistral-large-3, Harness: Arena text, style controlKimi K2.6: Style-controlled Arena rating from 9,562 anonymous pairwise votes.Mistral Large 3: Style-controlled Arena rating from 19,002 anonymous pairwise votes.	Winner1,440.989 95% confidence interval: ±7.079The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	1,388.294 95% confidence interval: ±4.988The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.	LMArena Source details Data as of: 09/25/2026