Skip to main content

Head-to-head comparison

GPT-5.6 Luna vs. DeepSeek-V4-Flash

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

shared result series
61
distinct benchmarks
2
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 GPT-5.6 Luna and DeepSeek-V4-Flash
AttributeGPT-5.6 LunaDeepSeek-V4-Flash
ProviderOpenAIDeepSeek
ReleasedJun 26, 2026Apr 24, 2026
AvailabilityActivePreview
Model typeProprietaryOpen Weights
ParametersNot published284B, 13B active
ArchitectureNot publishedMixture of Experts
Context window1,050,000 tokens1,000,000 tokens
Knowledge cutoffFebruary 16, 2026Not published
Technical sourcesModel, Context, Knowledge cutoffModel

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 GPT-5.6 Luna and DeepSeek-V4-Flash
Price typeGPT-5.6 LunaDeepSeek-V4-Flash
API input$0.2 ≤272K / $0.4 >272K$0.14
API output$1.2 ≤272K / $1.8 >272K$0.28
Cache read$0.02 ≤272K / $0.04 >272K$0.
Cache write (5 min.)$0.25 ≤272K / $0.5 >272KNot listed
Cache write (1 hr.)Not listedNot listed
Price verified08/11/2026 Source07/30/2026 Source
GPT-5.6 LunaDeepSeek-V4-Flash
API inputUSD per 1M tokens, base tier
GPT-5.6 Luna$0.2
DeepSeek-V4-FlashLower price$0.14
API outputUSD per 1M tokens, base tier
GPT-5.6 Luna$1.2
DeepSeek-V4-FlashLower price$0.28
Cache readUSD per 1M tokens, base tier
GPT-5.6 Luna$0.02
DeepSeek-V4-FlashLower price$0.003

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 GPT-5.6 Luna and DeepSeek-V4-Flash
ModelProviderPurchase route and tariffInput per 1M tokensOutput per 1M tokensSource
GPT-5.6 LunaOpenAIDirect from developerUp to 272,000 context tokens$0.2$1.2Source08/11/2026
OpenAIDirect from developerAbove 272,000 context tokens$0.4$1.8Source08/11/2026
Amazon BedrockVia OpenRouteramazon-bedrock/us-east-1$0.22$1.32Source09/07/2026
AzureVia OpenRouterazure$0.2$1.2Source09/07/2026
AzureVia OpenRouterazure/eu$0.22$1.32Source09/07/2026
AzureVia OpenRouterazure/us$0.22$1.32Source09/07/2026
OpenAIVia OpenRouteropenai$0.2$1.2Source09/07/2026
OpenAIVia OpenRouteropenai/fast$0.4$2.4Source09/07/2026
OpenAIVia OpenRouteropenai/flexLowest listed price: $0.1Lowest listed price: $0.6Source09/07/2026
DeepSeek-V4-FlashDeepSeekDirect from developerStandard API$0.14$0.28Source07/30/2026
AlibabaVia OpenRouteralibaba$0.352$1.056Source09/07/2026
AtlasCloudVia OpenRouteratlas-cloud/fp4$0.44$1.32Source09/07/2026
BaiduVia OpenRouterbaidu/fp8$0.14$0.28Source09/07/2026
BaseTenVia OpenRouterbaseten/fp8#1$0.13$0.26Source09/07/2026
BaseTenVia OpenRouterbaseten/fp8#2$0.13$0.26Source09/07/2026
CloudflareVia OpenRoutercloudflare$0.44$1.32Source09/07/2026
CoreWeaveVia OpenRoutercoreweave/fp8$0.13$0.28Source09/07/2026
DeepInfraVia OpenRouterdeepinfra/fp8$0.06$0.18Source09/07/2026
DeepSeekVia OpenRouterdeepseek$0.22$0.66Source09/07/2026
DigitalOceanVia OpenRouterdigitalocean$0.08$0.252Source09/07/2026
FireworksVia OpenRouterfireworks$0.22$0.66Source09/07/2026
GMICloudVia OpenRoutergmicloud/fp8$0.352$1.056Source09/07/2026
InceptronVia OpenRouterinceptron/fp4$0.13$0.28Source09/07/2026
MakoraVia OpenRoutermakora$0.09$0.195Source09/07/2026
Mancer 2Via OpenRoutermancer/fp8$0.15$0.5Source09/07/2026
MorphVia OpenRoutermorph/bf16$0.1234$0.3475Source09/07/2026
NextBitVia OpenRouternextbit/fp8$0.4$1.2Source09/07/2026
NovitaVia OpenRouternovita/fp8$0.4092$1.2276Source09/07/2026
OpenInferenceVia OpenRouteropen-inference/fp8$0.05$0.16Source09/07/2026
ParasailVia OpenRouterparasail/fp8$0.14$0.28Source09/07/2026
PhalaVia OpenRouterphala$0.44$1.32Source09/07/2026
RekaVia OpenRouterreka/fp4$0.11$0.66Source09/07/2026
RelaceVia OpenRouterrelace/fp4Lowest listed price: $0.045Lowest listed price: $0.09Source09/07/2026
Sail ResearchVia OpenRoutersail-research/fp4$0.065$0.18Source09/07/2026
SiliconFlowVia OpenRoutersiliconflow/fp8$0.22$0.66Source09/07/2026
StreamLakeVia OpenRouterstreamlake/fp8$0.088$0.264Source09/07/2026
TogetherVia OpenRoutertogether$0.14$0.28Source09/07/2026
VeniceVia OpenRoutervenice$0.175$0.35Source09/07/2026
WaferVia OpenRouterwafer/fast$0.1$0.25Source09/07/2026

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.

Not enough directly matched categories for a reliable radar

Only 1 of at least 5 categories have comparable, direct model measurements. The benchmark tables below show the available evidence without filling missing axes.

GPT-5.6 LunaDeepSeek-V4-Flash
CyberBenchOverall. Higher is better. No documented setup difference
GPT-5.6 LunaWinner83.898 %
DeepSeek-V4-Flash66.716 %
LMArena Text, Style ControlOverall. Higher is better. Test setup not documented as identical
GPT-5.6 LunaHigher measured value1,452.364
DeepSeek-V4-Flash1,435.816
CyberBenchPatch. Higher is better. No documented setup difference
GPT-5.6 LunaWinner79.661 %
DeepSeek-V4-Flash72.414 %
LMArena Text, Style ControlOverall. Higher is better. Test setup not documented as identical
GPT-5.6 LunaHigher measured value1,452.364
DeepSeek-V4-Flash1,438.462
CyberBenchPoC. Higher is better. No documented setup difference
GPT-5.6 LunaWinner88.136 %
DeepSeek-V4-Flash61.017 %
LMArena Text, Style ControlBusiness, management and finance. Higher is better. Test setup not documented as identical
GPT-5.6 LunaHigher measured value1,461.709
DeepSeek-V4-Flash1,437.854
LMArena Text, Style ControlBusiness, management and finance. Higher is better. Test setup not documented as identical
GPT-5.6 LunaHigher measured value1,461.709
DeepSeek-V4-Flash1,434.968
LMArena Text, Style ControlChinese. Higher is better. Test setup not documented as identical
GPT-5.6 LunaHigher measured value1,474.185
DeepSeek-V4-Flash1,472.677
LMArena Text, Style ControlChinese. Higher is better. Test setup not documented as identical
GPT-5.6 Luna1,474.185
DeepSeek-V4-FlashHigher measured value1,479.62
LMArena Text, Style ControlCoding. Higher is better. Test setup not documented as identical
GPT-5.6 LunaHigher measured value1,497.069
DeepSeek-V4-Flash1,483.101
LMArena Text, Style ControlCoding. Higher is better. Test setup not documented as identical
GPT-5.6 LunaHigher measured value1,497.069
DeepSeek-V4-Flash1,479.449
LMArena Text, Style ControlCreative writing. Higher is better. Test setup not documented as identical
GPT-5.6 LunaHigher measured value1,408.317
DeepSeek-V4-Flash1,408.18

Key to the marks:better valuebetter, but not conclusivetie

Benchmark scores for GPT-5.6 Luna and DeepSeek-V4-Flash
Benchmark and taskGPT-5.6 LunaDeepSeek-V4-FlashSource
CyberBenchOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner83.898 %
66.716 %
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 betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 24,433 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 48,963 anonymous pairwise votes.
Higher measured value1,452.364
95% confidence interval: ±5.03The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,435.816
95% confidence interval: ±4.15The 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
CyberBenchPatch
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner79.661 %
72.414 %
Vals AI
Source details
Data as of: 08/03/2026
LMArena Text, Style ControlOverall
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 24,433 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 48,705 anonymous pairwise votes.
Higher measured value1,452.364
95% confidence interval: ±5.03The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,438.462
95% confidence interval: ±4.21The 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
CyberBenchPoC
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner88.136 %
61.017 %
Vals AI
Source details
Data as of: 08/03/2026
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 betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 4,691 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 9,911 anonymous pairwise votes.
Higher measured value1,461.709
95% confidence interval: ±9.371The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,437.854
95% confidence interval: ±7.128The 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
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 betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 4,691 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 10,060 anonymous pairwise votes.
Higher measured value1,461.709
95% confidence interval: ±9.371The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,434.968
95% confidence interval: ±7.142The 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
LMArena Text, Style ControlChinese
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 1,691 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,530 anonymous pairwise votes.
Higher measured value1,474.185
95% confidence interval: ±15.484The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,472.677
95% confidence interval: ±12.483The 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
LMArena Text, Style ControlChinese
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 1,691 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,509 anonymous pairwise votes.
1,474.185
95% confidence interval: ±15.484The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,479.62
95% confidence interval: ±12.459The 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
LMArena Text, Style ControlCoding
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 6,843 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 14,346 anonymous pairwise votes.
Higher measured value1,497.069
95% confidence interval: ±7.965The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,483.101
95% confidence interval: ±6.376The 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
LMArena Text, Style ControlCoding
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 6,843 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 14,310 anonymous pairwise votes.
Higher measured value1,497.069
95% confidence interval: ±7.965The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,479.449
95% confidence interval: ±6.409The 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
LMArena Text, Style ControlCreative writing
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 4,866 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 7,972 anonymous pairwise votes.
Higher measured value1,408.317
95% confidence interval: ±9.528The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,408.18
95% confidence interval: ±7.747The 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
LMArena Text, Style ControlCreative writing
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 4,866 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 8,029 anonymous pairwise votes.
1,408.317
95% confidence interval: ±9.528The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,408.903
95% confidence interval: ±7.781The 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: 09/01/2026
LMArena Text, Style ControlEnglish
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 10,147 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 21,931 anonymous pairwise votes.
Higher measured value1,465.717
95% confidence interval: ±6.77The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,445.287
95% confidence interval: ±5.383The 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
LMArena Text, Style ControlEnglish
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 10,147 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 21,747 anonymous pairwise votes.
Higher measured value1,465.717
95% confidence interval: ±6.77The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,449.894
95% confidence interval: ±5.421The 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
LMArena Text, Style ControlEntertainment, sports and media
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 6,255 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 10,423 anonymous pairwise votes.
Higher measured value1,421.389
95% confidence interval: ±8.549The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,405.349
95% confidence interval: ±7.122The 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
LMArena Text, Style ControlEntertainment, sports and media
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 6,255 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 10,655 anonymous pairwise votes.
Higher measured value1,421.389
95% confidence interval: ±8.549The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,406.152
95% confidence interval: ±7.06The 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
LMArena Text, Style ControlExcluding ties
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 18,424 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 36,983 anonymous pairwise votes.
Higher measured value1,454.697
95% confidence interval: ±6.557The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,431.271
95% confidence interval: ±5.416The 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
LMArena Text, Style ControlExcluding ties
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 18,424 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 37,039 anonymous pairwise votes.
Higher measured value1,454.697
95% confidence interval: ±6.557The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,435.7
95% confidence interval: ±5.462The 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
LMArena Text, Style ControlExpert prompts
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 2,703 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 4,904 anonymous pairwise votes.
Higher measured value1,487.391
95% confidence interval: ±11.906The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,458.425
95% confidence interval: ±9.147The 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
LMArena Text, Style ControlExpert prompts
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 2,703 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 4,907 anonymous pairwise votes.
Higher measured value1,487.391
95% confidence interval: ±11.906The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,467.744
95% confidence interval: ±9.211The 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: 09/01/2026
LMArena Text, Style ControlFrench
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 835 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,909 anonymous pairwise votes.
Higher measured value1,476.247
95% confidence interval: ±21.942The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,444.246
95% confidence interval: ±15.542The 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
LMArena Text, Style ControlFrench
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 835 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,872 anonymous pairwise votes.
Higher measured value1,476.247
95% confidence interval: ±21.942The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,457.512
95% confidence interval: ±15.76The 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
LMArena Text, Style ControlGerman
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 424 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 834 anonymous pairwise votes.
Higher measured value1,486.451
95% confidence interval: ±29.435The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,423.985
95% confidence interval: ±21.432The 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
LMArena Text, Style ControlGerman
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 424 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 881 anonymous pairwise votes.
Higher measured value1,486.451
95% confidence interval: ±29.435The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,434.001
95% confidence interval: ±20.992The 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
LMArena Text, Style ControlHard prompts
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 16,256 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 32,576 anonymous pairwise votes.
Higher measured value1,472.892
95% confidence interval: ±5.879The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,459.19
95% confidence interval: ±4.879The 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
LMArena Text, Style ControlHard prompts
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 16,256 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 32,479 anonymous pairwise votes.
Higher measured value1,472.892
95% confidence interval: ±5.879The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,458.492
95% confidence interval: ±4.934The 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
LMArena Text, Style ControlHard prompts, English
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 6,739 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 15,327 anonymous pairwise votes.
Higher measured value1,482.99
95% confidence interval: ±7.984The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,464.306
95% confidence interval: ±6.191The 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
LMArena Text, Style ControlHard prompts, English
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 6,739 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 15,135 anonymous pairwise votes.
Higher measured value1,482.99
95% confidence interval: ±7.984The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,465.194
95% confidence interval: ±6.261The 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
LMArena Text, Style ControlInstruction following
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 8,687 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 16,772 anonymous pairwise votes.
Higher measured value1,442.672
95% confidence interval: ±7.285The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,428.139
95% confidence interval: ±5.955The 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
LMArena Text, Style ControlInstruction following
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 8,687 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 16,813 anonymous pairwise votes.
Higher measured value1,442.672
95% confidence interval: ±7.285The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,433.908
95% confidence interval: ±6.024The 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
LMArena Text, Style ControlJapanese
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 339 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 633 anonymous pairwise votes.
Higher measured value1,448.32
95% confidence interval: ±34.335The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,397.389
95% confidence interval: ±25.351The 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
LMArena Text, Style ControlJapanese
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 339 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 600 anonymous pairwise votes.
Higher measured value1,448.32
95% confidence interval: ±34.335The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,422.217
95% confidence interval: ±25.99The 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
LMArena Text, Style ControlKorean
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 481 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 867 anonymous pairwise votes.
Higher measured value1,398.575
95% confidence interval: ±27.208The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,394.225
95% confidence interval: ±21.434The 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
LMArena Text, Style ControlKorean
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 481 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 909 anonymous pairwise votes.
Higher measured value1,398.575
95% confidence interval: ±27.208The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,374.943
95% confidence interval: ±21.551The 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
LMArena Text, Style ControlLegal and government
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 2,064 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 3,892 anonymous pairwise votes.
Higher measured value1,462.472
95% confidence interval: ±13.711The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,449.174
95% confidence interval: ±10.185The 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
LMArena Text, Style ControlLegal and government
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 2,064 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 3,872 anonymous pairwise votes.
Higher measured value1,462.472
95% confidence interval: ±13.711The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,459.091
95% confidence interval: ±10.242The 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
LMArena Text, Style ControlLife, physical and social science
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 3,971 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 7,776 anonymous pairwise votes.
1,455.82
95% confidence interval: ±9.876The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,458.608
95% confidence interval: ±7.61The 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
LMArena Text, Style ControlLife, physical and social science
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 3,971 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 7,968 anonymous pairwise votes.
1,455.82
95% confidence interval: ±9.876The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,460.882
95% confidence interval: ±7.588The 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
LMArena Text, Style ControlLonger queries
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 11,564 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 21,561 anonymous pairwise votes.
Higher measured value1,451.975
95% confidence interval: ±6.845The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,448.456
95% confidence interval: ±5.733The 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
LMArena Text, Style ControlLonger queries
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 11,564 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 21,664 anonymous pairwise votes.
Higher measured value1,451.975
95% confidence interval: ±6.845The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,449.036
95% confidence interval: ±5.731The 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
LMArena Text, Style ControlMath
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 1,078 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,516 anonymous pairwise votes.
Higher measured value1,477.348
95% confidence interval: ±17.909The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,425.091
95% confidence interval: ±12.209The 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
LMArena Text, Style ControlMath
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 1,078 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,459 anonymous pairwise votes.
Higher measured value1,477.348
95% confidence interval: ±17.909The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,441.301
95% confidence interval: ±12.311The 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
LMArena Text, Style ControlMathematical professions
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 1,258 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,746 anonymous pairwise votes.
Higher measured value1,477.124
95% confidence interval: ±17.262The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,431.608
95% confidence interval: ±11.935The 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
LMArena Text, Style ControlMathematical professions
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 1,258 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,607 anonymous pairwise votes.
Higher measured value1,477.124
95% confidence interval: ±17.262The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,443.348
95% confidence interval: ±12.313The 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
LMArena Text, Style ControlMedicine and healthcare
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 1,853 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 3,429 anonymous pairwise votes.
1,436.426
95% confidence interval: ±14.729The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,459.604
95% confidence interval: ±10.945The 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
LMArena Text, Style ControlMedicine and healthcare
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 1,853 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 3,495 anonymous pairwise votes.
1,436.426
95% confidence interval: ±14.729The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,462.945
95% confidence interval: ±10.866The 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
LMArena Text, Style ControlMulti-turn
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 4,094 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 8,820 anonymous pairwise votes.
Higher measured value1,458.732
95% confidence interval: ±9.928The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,452.356
95% confidence interval: ±7.462The 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
LMArena Text, Style ControlMulti-turn
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 4,094 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 8,674 anonymous pairwise votes.
Higher measured value1,458.732
95% confidence interval: ±9.928The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,445.683
95% confidence interval: ±7.508The 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
LMArena Text, Style ControlNon-English
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 14,283 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 27,030 anonymous pairwise votes.
Higher measured value1,435.463
95% confidence interval: ±6.095The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,423.067
95% confidence interval: ±5.056The 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
LMArena Text, Style ControlNon-English
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 14,283 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 26,956 anonymous pairwise votes.
Higher measured value1,435.463
95% confidence interval: ±6.095The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,422.651
95% confidence interval: ±5.123The 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
LMArena Text, Style ControlPolish
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 382 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,005 anonymous pairwise votes.
1,429.959
95% confidence interval: ±29.891The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,439.462
95% confidence interval: ±18.965The 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
LMArena Text, Style ControlPolish
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 382 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,014 anonymous pairwise votes.
1,429.959
95% confidence interval: ±29.891The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,430.358
95% confidence interval: ±18.866The 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
LMArena Text, Style ControlRussian
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 2,470 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 5,250 anonymous pairwise votes.
Higher measured value1,445.524
95% confidence interval: ±12.278The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,433.186
95% confidence interval: ±8.845The 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
LMArena Text, Style ControlRussian
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 2,470 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 5,207 anonymous pairwise votes.
Higher measured value1,445.524
95% confidence interval: ±12.278The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,437.38
95% confidence interval: ±8.983The 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
LMArena Text, Style ControlSoftware and IT services
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 9,767 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 20,105 anonymous pairwise votes.
Higher measured value1,491.879
95% confidence interval: ±6.977The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,471.075
95% confidence interval: ±5.634The 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
LMArena Text, Style ControlSoftware and IT services
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 9,767 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 20,031 anonymous pairwise votes.
Higher measured value1,491.879
95% confidence interval: ±6.977The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,473.269
95% confidence interval: ±5.664The 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
LMArena Text, Style ControlSpanish
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 666 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,548 anonymous pairwise votes.
Higher measured value1,462.579
95% confidence interval: ±24.039The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,433.026
95% confidence interval: ±16.368The 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
LMArena Text, Style ControlSpanish
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 666 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,443 anonymous pairwise votes.
Higher measured value1,462.579
95% confidence interval: ±24.039The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,431.772
95% confidence interval: ±16.888The 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
LMArena Text, Style ControlWriting, literature and language
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 6,482 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 11,742 anonymous pairwise votes.
Higher measured value1,434.128
95% confidence interval: ±8.276The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,417.982
95% confidence interval: ±6.712The 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
LMArena Text, Style ControlWriting, literature and language
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-5.6 Luna: Snapshot: gpt-5.6-luna-xhigh, Reasoning: xhigh, Harness: Arena text, style control
DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlGPT-5.6 Luna: Style-controlled Arena rating from 6,482 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 11,861 anonymous pairwise votes.
Higher measured value1,434.128
95% confidence interval: ±8.276The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
1,421.395
95% confidence interval: ±6.717The 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

Key to the marks:better valuebetter, but not conclusivetie

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

Related LLM comparisons

Compare GPT-5.6 Luna and DeepSeek-V4-Flash with other leading models using the same data and benchmark logic.

View the complete LLM comparison