Skip to main content

Head-to-head comparison

DeepSeek-V4-Pro vs. Kimi K3

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

shared result series
210
distinct benchmarks
22
latest retrieval
2026-09-04

Change model selection

Specifications

General specifications

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

Specifications for DeepSeek-V4-Pro and Kimi K3
AttributeDeepSeek-V4-ProKimi K3
ProviderDeepSeekMoonshot AI
ReleasedApr 24, 2026Jul 16, 2026
AvailabilityPreviewActive
Model typeOpen WeightsOpen Weights
Parameters1.6T, 49B active2.8T, 104B active
ArchitectureMixture of ExpertsMixture of Experts
Context window1,000,000 tokens1,048,576 tokens
Knowledge cutoffNot publishedNot published
Technical sourcesModel, ContextModel

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 DeepSeek-V4-Pro and Kimi K3
Price typeDeepSeek-V4-ProKimi K3
API input$0.66$3
API output$1.98$15
Cache read$0.02$0.3
Cache write (5 min.)Not listedNot listed
Cache write (1 hr.)Not listedNot listed
Price verified09/08/2026 Source07/30/2026 Source
DeepSeek-V4-ProKimi K3
API inputUSD per 1M tokens, base tier
DeepSeek-V4-ProLower price$0.66
Kimi K3$3
API outputUSD per 1M tokens, base tier
DeepSeek-V4-ProLower price$1.98
Kimi K3$15
Cache readUSD per 1M tokens, base tier
DeepSeek-V4-ProLower price$0.022
Kimi K3$0.3

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 DeepSeek-V4-Pro and Kimi K3
ModelProviderPurchase route and tariffInput per 1M tokensOutput per 1M tokensSource
DeepSeek-V4-ProDeepSeekDirect from developerStandard API$0.66$1.98Source09/08/2026
AlibabaVia OpenRouteralibaba$1.122$3.366Source09/12/2026
BaiduVia OpenRouterbaidu/fp8Lowest listed price: $0.5782Lowest listed price: $1.7345Source09/12/2026
BaseTenVia OpenRouterbaseten/fp4#1$1.32$3.96Source09/12/2026
BaseTenVia OpenRouterbaseten/fp4#2$1.32$3.96Source09/12/2026
CloudflareVia OpenRoutercloudflare$1.32$3.96Source09/12/2026
CoreWeaveVia OpenRoutercoreweave/fp8$1.31$3.96Source09/12/2026
DeepInfraVia OpenRouterdeepinfra/fp8$1.3$2.6Source09/12/2026
DeepSeekVia OpenRouterdeepseek$0.66$1.98Source09/12/2026
DigitalOceanVia OpenRouterdigitalocean$1.32$3.96Source09/12/2026
FireworksVia OpenRouterfireworks$1.32$3.96Source09/12/2026
GMICloudVia OpenRoutergmicloud/fp8$1.056$3.168Source09/12/2026
IonstreamVia OpenRouterionstream$0.88$2.64Source09/12/2026
NextBitVia OpenRouternextbit/fp8$1.122$3.366Source09/12/2026
NovitaVia OpenRouternovita/fp8$0.99$2.97Source09/12/2026
ParasailVia OpenRouterparasail/fp8$1.32$3.96Source09/12/2026
PhalaVia OpenRouterphala$1.45$4.36Source09/12/2026
Sail ResearchVia OpenRoutersail-research/fp4$1.32$3.96Source09/12/2026
SiliconFlowVia OpenRoutersiliconflow/fp8$1.32$3.96Source09/12/2026
StreamLakeVia OpenRouterstreamlake$1.0547$3.164Source09/12/2026
TogetherVia OpenRoutertogether$1.32$3.96Source09/12/2026
VeniceVia OpenRoutervenice$1.65$4.95Source09/12/2026
Kimi K3Moonshot AIDirect from developerStandard API$3$15Source07/30/2026
AlibabaVia OpenRouteralibaba$3.45$17.25Source09/12/2026
BaseTenVia OpenRouterbaseten/fp8$3$15Source09/12/2026
ChutesVia OpenRouterchutes/mxfp4$3$15Source09/12/2026
DeepInfraVia OpenRouterdeepinfra/bf16$2.85$14.25Source09/12/2026
DigitalOceanVia OpenRouterdigitalocean$2.55$12.95Source09/12/2026
FireworksVia OpenRouterfireworks$3$15Source09/12/2026
FireworksVia OpenRouterfireworks/fast$4.5$22.5Source09/12/2026
FireworksVia OpenRouterfireworks/us$3.3$16.5Source09/12/2026
MakoraVia OpenRoutermakora$2.55$12.75Source09/12/2026
ModalVia OpenRoutermodal/mxfp4$3$15Source09/12/2026
Moonshot AIVia OpenRoutermoonshotai/mxfp4$3$15Source09/12/2026
MorphVia OpenRoutermorph/fast$6$22.5Source09/12/2026
MorphVia OpenRoutermorph/fp8Lowest listed price: $2.125$11.9Source09/12/2026
ParasailVia OpenRouterparasail/fp4$3$15Source09/12/2026
PhalaVia OpenRouterphala$2.85$14.25Source09/12/2026
RelaceVia OpenRouterrelace/fp4$2.4$12Source09/12/2026
Sail ResearchVia OpenRoutersail-research/fp4$2.3027Lowest listed price: $11.5502Source09/12/2026
TogetherVia OpenRoutertogether$3$15Source09/12/2026
WaferVia OpenRouterwafer$3$12.75Source09/12/2026

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.

255075100CodingReasoningMathematicsKnowledgeFinanceLegalAgentic tasksProductivity
DeepSeek-V4-Pro
Kimi K3
Coding: 68.3 / 92.2 (5 benchmark families)
Reasoning: 96.2 / 100 (1 benchmark family)
Mathematics: 18.4 / 100 (1 benchmark family)
Knowledge: 99.2 / 100 (1 benchmark family)
Finance: 85 / 100 (4 benchmark families)
Legal: 60.1 / 100 (3 benchmark families)
Agentic tasks: 62 / 100 (1 benchmark family)
Productivity: 92.2 / 100 (1 benchmark family)

22 matched rows from 21 benchmark families were considered. The radar shows 8 of 11 comparable categories.

DeepSeek-V4-ProKimi K3
Code MigrationOverall. Higher is better. No documented setup difference
DeepSeek-V4-ProWinner26.201 %
Kimi K316.099 %
CorpFin v2Overall. Higher is better. No documented setup difference
DeepSeek-V4-Pro61.383 %
Kimi K3Winner71.562 %
CyberBenchOverall. Higher is better. No documented setup difference
DeepSeek-V4-Pro68.644 %
Kimi K3Winner79.026 %
EMBOverall. Higher is better. No documented setup difference
DeepSeek-V4-Pro51.623 %
Kimi K3Winner66.399 %
Finance Agent (v2)Overall. Higher is better. No documented setup difference
DeepSeek-V4-Pro44.083 %
Kimi K3Winner54.36 %
GPQA DiamondOverall. Higher is better. No documented setup difference
DeepSeek-V4-Pro89.394 %
Kimi K3Winner92.929 %
Harvey's Legal Agent BenchmarkOverall · Task fully resolved. Higher is better. No documented setup difference
DeepSeek-V4-Pro3.75 %
Kimi K3Winner10.833 %
Legal Research BenchOverall · All-pass. Higher is better. No documented setup difference
DeepSeek-V4-Pro23.077 %
Kimi K3Winner44.231 %
LegalBenchOverall. Higher is better. No documented setup difference
DeepSeek-V4-Pro80.323 %
Kimi K3Winner86.016 %
LiveCodeBench v6Overall. Higher is better. No documented setup difference
DeepSeek-V4-ProWinner87.484 %
Kimi K387.185 %
LMArena Text, Style ControlOverall. Higher is better. Test setup not documented as identical
DeepSeek-V4-Pro1,457.748
Kimi K3Higher measured value1,489.017
MedCodeOverall. Higher is better. No documented setup difference
DeepSeek-V4-Pro40.455 %
Kimi K3Winner48.884 %

Key to the marks:better valuebetter, but not conclusivetie

Benchmark scores for DeepSeek-V4-Pro and Kimi K3
Benchmark and taskDeepSeek-V4-ProKimi K3Source
Code MigrationOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner26.201 %
16.099 %
Vals AI
Source details
Data as of: 09/03/2026
CorpFin v2Overall
Test details
No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is better
61.383 %
Winner71.562 %
Vals AI
Source details
Data as of: 08/12/2026
CyberBenchOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
68.644 %
Winner79.026 %
Vals AI
Source details
Data as of: 09/01/2026
EMBOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
51.623 %
Winner66.399 %
Vals AI
Source details
Data as of: 09/03/2026
Finance Agent (v2)Overall
Test details
No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is better
44.083 %
Winner54.36 %
Vals AI
Source details
Data as of: 09/03/2026
GPQA DiamondOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
89.394 %
Winner92.929 %
Vals AI
Source details
Data as of: 09/01/2026
Harvey's Legal Agent BenchmarkOverall · Task fully resolved
Test details
No documented difference in the test setupVersion: 1Metric: task resolution rateScoring: Higher is better
3.75 %
Winner10.833 %
Vals AI
Source details
Data as of: 09/03/2026
Legal Research BenchOverall · All-pass
Test details
No documented difference in the test setupVersion: 1Metric: all-pass rateScoring: Higher is better
23.077 %
Winner44.231 %
Vals AI
Source details
Data as of: 09/03/2026
LegalBenchOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
80.323 %
Winner86.016 %
Vals AI
Source details
Data as of: 09/01/2026
LiveCodeBench v6Overall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner87.484 %
87.185 %
Vals AI
Source details
Data as of: 09/01/2026
LMArena Text, Style ControlOverall
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control
Kimi K3: Snapshot: kimi-k3-max, Reasoning: max, Harness: Arena text, style control
DeepSeek-V4-Pro: Style-controlled Arena rating from 54,243 anonymous pairwise votes.Kimi K3: Style-controlled Arena rating from 17,895 anonymous pairwise votes.
1,457.748
95% confidence interval: ±4.009The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,489.017
95% confidence interval: ±5.47The 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: 09/01/2026
MedCodeOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
40.455 %
Winner48.884 %
Vals AI
Source details
Data as of: 09/03/2026
MedScribeOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
75.144 %
Winner87.958 %
Vals AI
Source details
Data as of: 09/03/2026
MMLU-ProOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
87.249 %
Winner87.973 %
Vals AI
Source details
Data as of: 09/01/2026
ProgramBenchOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
0 %
Winner2 %
Vals AI
Source details
Data as of: 09/01/2026
ProofBenchOverall
Test details
No documented difference in the test setupVersion: 1.1Metric: accuracyScoring: Higher is better
16 %
Winner87 %
Vals AI
Source details
Data as of: 09/01/2026
Public Benefits Bench v1.1Overall
Test details
No documented difference in the test setupVersion: 1.1Metric: accuracyScoring: Higher is better
62.923 %
Winner68.268 %
Vals AI
Source details
Data as of: 09/01/2026
SWE-bench VerifiedOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
77.4 %
Winner93.4 %
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
72.077 %
Winner75.716 %
Vals AI
Source details
Data as of: 09/01/2026
Terminal-Bench 2.1Overall
Test details
No documented difference in the test setupVersion: 2.1Metric: accuracyScoring: Higher is better
50.187 %
Winner80.899 %
Vals AI
Source details
Data as of: 09/03/2026
Vals IndexOverall
Test details
No documented difference in the test setupVersion: 2Metric: weighted index scoreScoring: Higher is better
42.888 %
Winner57.813 %
Vals AI
Source details
Data as of: 09/03/2026
Vibe Code Bench v1.1Overall
Test details
No documented difference in the test setupVersion: 1.1Metric: accuracyScoring: Higher is betterSettings: Harness: OpenHands
49.931 %
Winner84.963 %
Vals AI
Source details
Data as of: 09/03/2026
CorpFin v2Exact Pages
Test details
Test setup not documented as identicalNo direct winnerVersion: 2Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
59.207 %
Higher measured value69.347 %
Vals AI
Source details
Data as of: 08/06/2026
CyberBenchPatch
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterKimi K3: Temperature: 1, Output limit: 262,144 tokens
79.661 %
Higher measured value83.051 %
Vals AI
Source details
Data as of: 08/03/2026
EMBComps
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
57.434 %
Higher measured value72.685 %
Vals AI
Source details
Data as of: 08/06/2026
Finance Agent (v2)Adjustments
Test details
Test setup not documented as identicalNo direct winnerVersion: 2Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
41.586 %
Higher measured value51.461 %
Vals AI
Source details
Data as of: 08/06/2026
GPQA DiamondFew-Shot CoT
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
90.404 %
Higher measured value93.434 %
Vals AI
Source details
Data as of: 08/09/2026
Harvey's Legal Agent BenchmarkAntitrust Competition · Criteria passed
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: criteria pass rateScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
83.696 %
Higher measured value87.319 %
Vals AI
Source details
Data as of: 08/07/2026
Legal Research BenchAdministrative / Regulatory · All-pass
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: all-pass rateScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
28.788 %
Higher measured value54.545 %
Vals AI
Source details
Data as of: 08/06/2026
LegalBenchConclusion Tasks
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
89.117 %
Higher measured value90.733 %
Vals AI
Source details
Data as of: 08/09/2026
LiveCodeBench v6Easy
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
No direct winner98.758 %
No direct winner98.758 %
Vals AI
Source details
Data as of: 08/09/2026
LMArena Text, Style ControlOverall
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control
Kimi K3: Snapshot: kimi-k3-max, Reasoning: max, Harness: Arena text, style control
DeepSeek-V4-Pro: Style-controlled Arena rating from 51,599 anonymous pairwise votes.Kimi K3: Style-controlled Arena rating from 17,895 anonymous pairwise votes.
1,455.047
95% confidence interval: ±4.08The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,489.017
95% confidence interval: ±5.47The 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: 09/01/2026
MMLU-ProBiology
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
92.887 %
Higher measured value93.863 %
Vals AI
Source details
Data as of: 08/09/2026
ProgramBenchAlmost Resolved
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
0 %
Higher measured value17.5 %
Vals AI
Source details
Data as of: 08/09/2026
SWE-bench Verified<15 min fix
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
87.629 %
Higher measured value92.268 %
Vals AI
Source details
Data as of: 08/08/2026
TaxEval v2Correctness
Test details
Test setup not documented as identicalNo direct winnerVersion: 2Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
60.589 %
Higher measured value65.331 %
Vals AI
Source details
Data as of: 08/09/2026
Terminal-Bench 2.1Easy
Test details
Test setup not documented as identicalNo direct winnerVersion: 2.1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
66.667 %
Higher measured value91.667 %
Vals AI
Source details
Data as of: 08/06/2026
Vals IndexCorpFin v2
Test details
Test setup not documented as identicalNo direct winnerVersion: 1.2Metric: benchmark scoreScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
61.888 %
Higher measured value72.611 %
Vals AI
Source details
Data as of: 08/09/2026
CorpFin v2Max Fitting Context
Test details
Test setup not documented as identicalNo direct winnerVersion: 2Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
63.054 %
Higher measured value72.727 %
Vals AI
Source details
Data as of: 08/06/2026
CyberBenchPoC
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterKimi K3: Temperature: 1, Output limit: 262,144 tokens
57.627 %
Higher measured value75 %
Vals AI
Source details
Data as of: 08/03/2026
EMBDataroom
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
62.572 %
Higher measured value85.13 %
Vals AI
Source details
Data as of: 08/06/2026
Finance Agent (v2)All-Pass
Test details
Test setup not documented as identicalNo direct winnerVersion: 2Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
31.448 %
Higher measured value41.69 %
Vals AI
Source details
Data as of: 08/06/2026
GPQA DiamondZero-Shot CoT
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
88.384 %
Higher measured value92.424 %
Vals AI
Source details
Data as of: 08/09/2026
Harvey's Legal Agent BenchmarkAntitrust Competition · Task fully resolved
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: task resolution rateScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
No direct winner0 %
No direct winner0 %
Vals AI
Source details
Data as of: 08/07/2026
Legal Research BenchAdministrative / Regulatory · Weighted partial credit
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: weighted partial-credit scoreScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
83.372 %
Higher measured value90.731 %
Vals AI
Source details
Data as of: 08/06/2026
LegalBenchInterpretation Tasks
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
75.169 %
Higher measured value82.612 %
Vals AI
Source details
Data as of: 08/09/2026
LiveCodeBench v6Hard
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
72.571 %
Higher measured value74.286 %
Vals AI
Source details
Data as of: 08/09/2026
LMArena Text, Style ControlBusiness, management and finance
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control
Kimi K3: Snapshot: kimi-k3-max, Reasoning: max, Harness: Arena text, style control
DeepSeek-V4-Pro: Style-controlled Arena rating from 10,742 anonymous pairwise votes.Kimi K3: Style-controlled Arena rating from 3,520 anonymous pairwise votes.
1,461.034
95% confidence interval: ±6.908The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,489.216
95% confidence interval: ±10.41The 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: 09/01/2026
MMLU-ProBusiness
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
91.888 %
Higher measured value92.015 %
Vals AI
Source details
Data as of: 08/09/2026
ProgramBenchFully Resolved
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
0 %
Higher measured value2 %
Vals AI
Source details
Data as of: 08/09/2026
SWE-bench Verified>4 hours
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
0 %
Higher measured value66.667 %
Vals AI
Source details
Data as of: 08/08/2026
TaxEval v2Stepwise Reasoning
Test details
Test setup not documented as identicalNo direct winnerVersion: 2Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
83.565 %
Higher measured value86.1 %
Vals AI
Source details
Data as of: 08/09/2026
Terminal-Bench 2.1Hard
Test details
Test setup not documented as identicalNo direct winnerVersion: 2.1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
31.111 %
Higher measured value74.444 %
Vals AI
Source details
Data as of: 08/06/2026
Vals IndexFinance Agent v2
Test details
Test setup not documented as identicalNo direct winnerVersion: 1.2Metric: benchmark scoreScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
44.081 %
Higher measured value55.882 %
Vals AI
Source details
Data as of: 08/09/2026
CorpFin v2Shared Max Context
Test details
Test setup not documented as identicalNo direct winnerVersion: 2Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
61.888 %
Higher measured value72.611 %
Vals AI
Source details
Data as of: 08/06/2026
EMBDCF
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
44.435 %
Higher measured value58.877 %
Vals AI
Source details
Data as of: 08/06/2026
Finance Agent (v2)Comparables
Test details
Test setup not documented as identicalNo direct winnerVersion: 2Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
39.632 %
Higher measured value46.691 %
Vals AI
Source details
Data as of: 08/06/2026
Harvey's Legal Agent BenchmarkArbitration International Dispute Resolution · Criteria passed
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: criteria pass rateScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
83.512 %
Higher measured value92.115 %
Vals AI
Source details
Data as of: 08/07/2026
Legal Research BenchBusiness & Commercial · All-pass
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: all-pass rateScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
23.256 %
Higher measured value46.512 %
Vals AI
Source details
Data as of: 08/06/2026
LegalBenchIssue Tasks
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
84.587 %
Higher measured value91.099 %
Vals AI
Source details
Data as of: 08/09/2026
LiveCodeBench v6Medium
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 128,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
Higher measured value91.123 %
88.512 %
Vals AI
Source details
Data as of: 08/09/2026
LMArena Text, Style ControlBusiness, management and finance
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control
Kimi K3: Snapshot: kimi-k3-max, Reasoning: max, Harness: Arena text, style control
DeepSeek-V4-Pro: Style-controlled Arena rating from 10,411 anonymous pairwise votes.Kimi K3: Style-controlled Arena rating from 3,520 anonymous pairwise votes.
1,448.251
95% confidence interval: ±6.986The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,489.216
95% confidence interval: ±10.41The 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: 09/01/2026
MMLU-ProChemistry
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 256,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
Higher measured value92.491 %
91.784 %
Vals AI
Source details
Data as of: 08/09/2026
ProgramBenchRaw Pass Rate
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterDeepSeek-V4-Pro: Reasoning: max, Output limit: 384,000 tokens
Kimi K3: Temperature: 1, Output limit: 262,144 tokens
47.778 %
Higher measured value62.768 %
Vals AI
Source details
Data as of: 08/09/2026

Key to the marks:better valuebetter, but not conclusivetie

Our model tests

Community tests you can inspect

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.

DeepSeek-V4-Pro

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

First attempt
Run date
7 Sept 2026
API provider
openrouter
Requested API model
deepseek/deepseek-v4-pro-0813
Pinned endpoint
DeepSeek | deepseek/deepseek-v4-pro-20260813
Reasoning setting
high
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.

Kimi K3

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

First attempt
Run date
7 Sept 2026
API provider
openrouter
Requested API model
moonshotai/kimi-k3
Pinned endpoint
Moonshot AI | moonshotai/kimi-k3-20260715
Reasoning setting
high
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 DeepSeek-V4-Pro and Kimi K3 with other leading models using the same data and benchmark logic.

View the complete LLM comparison