Head-to-head comparison
DeepSeek-V4-Pro 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.
61
shared result series
2
distinct benchmarks
2026-08-16
latest retrieval
Change model selection
Specifications and cost
General specifications and standard API pricing
Undisclosed parameter counts and knowledge cutoffs stay marked as unavailable. The table shows direct standard token pricing, including documented context tiers. Batch, fast, priority, regional surcharges, tool calls, and cloud platform pricing are excluded.
| Attribute | DeepSeek-V4-Pro | DeepSeek-V4-Flash |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Released | Apr 24, 2026 | Apr 24, 2026 |
| Availability | Preview | Preview |
| Model type | Open Weights | Open Weights |
| Parameters | 1.6T, 49B active | 284B, 13B active |
| Architecture | Mixture of Experts | Mixture of Experts |
| Context window | 1,000,000 tokens | 1,000,000 tokens |
| Knowledge cutoff | Not published | Not published |
| Technical sources | Model, Context | Model |
| API pricing per 1M tokens | ||
| API input | $0.43 | $0.14 |
| API output | $0.87 | $0.28 |
| Cache read | $0. | $0. |
| Cache write (5 min.) | Not listed | Not listed |
| Cache write (1 hr.) | Not listed | Not listed |
| Price verified | 07/30/2026Source | 07/30/2026Source |
DeepSeek-V4-ProDeepSeek-V4-Flash
API inputUSD per 1M tokens, base tier
DeepSeek-V4-Pro$0.435
DeepSeek-V4-Flash$0.14Lower price
API outputUSD per 1M tokens, base tier
DeepSeek-V4-Pro$0.87
DeepSeek-V4-Flash$0.28Lower price
Cache readUSD per 1M tokens, base tier
DeepSeek-V4-Pro$0.004
DeepSeek-V4-Flash$0.003Lower price
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.
61 shared result series
DeepSeek-V4-ProDeepSeek-V4-Flash
Arena Text, Style ControlOverall. Higher is better. No documented setup difference
DeepSeek-V4-Pro1,458.103 arena pointsWinner
DeepSeek-V4-Flash1,435.476 arena points
CyberBenchOverall. Higher is better. No documented setup difference
DeepSeek-V4-Pro68.644 %Winner
DeepSeek-V4-Flash66.716 %
Arena Text, Style ControlOverall. Higher is better. No documented setup difference
DeepSeek-V4-Pro1,455.61 arena pointsWinner
DeepSeek-V4-Flash1,438.304 arena points
CyberBenchPatch. Higher is better. No documented setup difference
DeepSeek-V4-Pro79.661 %Winner
DeepSeek-V4-Flash72.414 %
Arena Text, Style ControlBusiness, management and finance. Higher is better. No documented setup difference
DeepSeek-V4-Pro1,460.638 arena pointsWinner
DeepSeek-V4-Flash1,437.217 arena points
CyberBenchPoC. Higher is better. No documented setup difference
DeepSeek-V4-Pro57.627 %
DeepSeek-V4-Flash61.017 %Winner
Arena Text, Style ControlBusiness, management and finance. Higher is better. No documented setup difference
DeepSeek-V4-Pro1,447.754 arena pointsNo clear lead
DeepSeek-V4-Flash1,435.23 arena points
Arena Text, Style ControlChinese. Higher is better. No documented setup difference
DeepSeek-V4-Pro1,494.914 arena pointsNo clear lead
DeepSeek-V4-Flash1,474.544 arena points
Arena Text, Style ControlChinese. Higher is better. No documented setup difference
DeepSeek-V4-Pro1,500.632 arena pointsNo clear lead
DeepSeek-V4-Flash1,481.177 arena points
Arena Text, Style ControlCoding. Higher is better. No documented setup difference
DeepSeek-V4-Pro1,502.053 arena pointsWinner
DeepSeek-V4-Flash1,483.133 arena points
Arena Text, Style ControlCoding. Higher is better. No documented setup difference
DeepSeek-V4-Pro1,488.583 arena pointsNo clear lead
DeepSeek-V4-Flash1,479.521 arena points
Arena Text, Style ControlCreative writing. Higher is better. No documented setup difference
DeepSeek-V4-Pro1,445.32 arena pointsWinner
DeepSeek-V4-Flash1,407.934 arena points
| Benchmark and task | DeepSeek-V4-Pro | DeepSeek-V4-Flash | Comparability | Source |
|---|---|---|---|---|
| Arena Text, Style ControlOverall | 1,458.103 arena pointsWinner 95% confidence interval: ±4.022The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,435.476 arena points 95% confidence interval: ±4.153The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:overallNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 54,288 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 49,096 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| CyberBenchOverall | 68.644 %Winner | 66.716 % | 1accuracyHigher is betterCohort: vals-cyberbench:1:overallNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: No further settings disclosed DeepSeek-V4-Flash: No further settings disclosed | Vals AIType: Independent evaluationLeaderboard updated: 2026-08-03Import ID: vals-cyber-2026-08-16-702acaee9aa9Retrieved 2026-08-16 |
| Arena Text, Style ControlOverallReasoning: high | 1,455.61 arena pointsWinner 95% confidence interval: ±4.089The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,438.304 arena points 95% confidence interval: ±4.22The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:overallNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 51,670 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 48,789 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| CyberBenchPatch | 79.661 %Winner | 72.414 % | 1accuracyHigher is betterCohort: vals-cyberbench:1:patchNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: No further settings disclosed DeepSeek-V4-Flash: No further settings disclosed | Vals AIType: Independent evaluationLeaderboard updated: 2026-08-03Import ID: vals-cyber-2026-08-11-1f20498a3dc1Retrieved 2026-08-11 |
| Arena Text, Style ControlBusiness, management and finance | 1,460.638 arena pointsWinner 95% confidence interval: ±6.951The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,437.217 arena points 95% confidence interval: ±7.157The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_business_and_management_and_financial_operationsNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 10,715 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 9,916 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| CyberBenchPoC | 57.627 % | 61.017 %Winner | 1accuracyHigher is betterCohort: vals-cyberbench:1:pocNo documented setup differenceWinner: DeepSeek-V4-FlashDeepSeek-V4-Pro: No further settings disclosed DeepSeek-V4-Flash: No further settings disclosed | Vals AIType: Independent evaluationLeaderboard updated: 2026-08-03Import ID: vals-cyber-2026-08-11-1f20498a3dc1Retrieved 2026-08-11 |
| Arena Text, Style ControlBusiness, management and financeReasoning: high | 1,447.754 arena pointsNo clear lead 95% confidence interval: ±7.037The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,435.23 arena points 95% confidence interval: ±7.178The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_business_and_management_and_financial_operationsNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 10,363 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 10,055 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlChinese | 1,494.914 arena pointsNo clear lead 95% confidence interval: ±12.109The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,474.544 arena points 95% confidence interval: ±12.686The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:chineseNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 2,881 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,557 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlChineseReasoning: high | 1,500.632 arena pointsNo clear lead 95% confidence interval: ±12.361The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,481.177 arena points 95% confidence interval: ±12.66The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:chineseNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 2,780 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,529 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlCoding | 1,502.053 arena pointsWinner 95% confidence interval: ±6.098The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,483.133 arena points 95% confidence interval: ±6.395The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:codingNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 16,219 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 14,397 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlCodingReasoning: high | 1,488.583 arena pointsNo clear lead 95% confidence interval: ±6.254The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,479.521 arena points 95% confidence interval: ±6.432The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:codingNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 15,137 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 14,353 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlCreative writing | 1,445.32 arena pointsWinner 95% confidence interval: ±7.549The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,407.934 arena points 95% confidence interval: ±7.787The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:creative_writingNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 8,993 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 7,974 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlCreative writingReasoning: high | 1,440.915 arena pointsWinner 95% confidence interval: ±7.633The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,408.648 arena points 95% confidence interval: ±7.808The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:creative_writingNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 8,630 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 8,019 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlEnglish | 1,467.841 arena pointsWinner 95% confidence interval: ±5.232The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,445.34 arena points 95% confidence interval: ±5.378The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:englishNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 23,938 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 21,994 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlEnglishReasoning: high | 1,466.432 arena pointsWinner 95% confidence interval: ±5.282The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,450.015 arena points 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. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:englishNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 22,716 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 21,774 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlEntertainment, sports and media | 1,434.546 arena pointsWinner 95% confidence interval: ±6.838The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,405.522 arena points 95% confidence interval: ±7.152The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_entertainment_and_sports_and_mediaNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 11,864 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 10,453 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlEntertainment, sports and mediaReasoning: high | 1,430.998 arena pointsWinner 95% confidence interval: ±6.917The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,405.113 arena points 95% confidence interval: ±7.086The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_entertainment_and_sports_and_mediaNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 11,417 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 10,654 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlExcluding ties | 1,461.846 arena pointsWinner 95% confidence interval: ±5.269The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,430.81 arena points 95% confidence interval: ±5.433The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:exclude_tiesNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 41,089 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 37,067 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlExcluding tiesReasoning: high | 1,457.694 arena pointsWinner 95% confidence interval: ±5.283The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,435.522 arena points 95% confidence interval: ±5.476The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:exclude_tiesNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 39,206 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 37,085 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlExpert prompts | 1,482.281 arena pointsWinner 95% confidence interval: ±8.872The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,458.397 arena points 95% confidence interval: ±9.185The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:expertNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 5,438 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 4,926 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlExpert promptsReasoning: high | 1,474.832 arena pointsNo clear lead 95% confidence interval: ±9.032The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,467.529 arena points 95% confidence interval: ±9.251The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:expertNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 5,066 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 4,924 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlFrench | 1,483.706 arena pointsWinner 95% confidence interval: ±15.355The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,445.102 arena points 95% confidence interval: ±15.765The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:frenchNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 2,116 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,916 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlFrenchReasoning: high | 1,470.225 arena pointsNo clear lead 95% confidence interval: ±15.468The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,458.295 arena points 95% confidence interval: ±15.946The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:frenchNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 1,986 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,884 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlGerman | 1,465.944 arena pointsWinner 95% confidence interval: ±20.28The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,423.659 arena points 95% confidence interval: ±21.582The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:germanNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 950 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 833 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlGermanReasoning: high | 1,458.46 arena pointsNo clear lead 95% confidence interval: ±20.895The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,435.281 arena points 95% confidence interval: ±21.205The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:germanNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 877 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 872 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlHard prompts | 1,480.536 arena pointsWinner 95% confidence interval: ±4.713The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,459.153 arena points 95% confidence interval: ±4.892The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:hard_promptsNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 36,125 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 32,675 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlHard promptsReasoning: high | 1,474.247 arena pointsWinner 95% confidence interval: ±4.773The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,458.18 arena points 95% confidence interval: ±4.95The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:hard_promptsNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 34,588 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 32,535 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlHard prompts, English | 1,484.514 arena pointsWinner 95% confidence interval: ±6.003The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,464.017 arena points 95% confidence interval: ±6.178The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:hard_prompts_englishNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 16,817 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 15,382 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlHard prompts, EnglishReasoning: high | 1,482.312 arena pointsWinner 95% confidence interval: ±6.088The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,464.963 arena points 95% confidence interval: ±6.257The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:hard_prompts_englishNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 15,915 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 15,145 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlInstruction following | 1,453.655 arena pointsWinner 95% confidence interval: ±5.745The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,428.017 arena points 95% confidence interval: ±5.969The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:instruction_followingNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 18,719 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 16,832 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlInstruction followingReasoning: high | 1,449.35 arena pointsWinner 95% confidence interval: ±5.839The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,434.165 arena points 95% confidence interval: ±6.039The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:instruction_followingNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 17,752 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 16,841 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlJapanese | 1,461.924 arena pointsWinner 95% confidence interval: ±25.076The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,396.126 arena points 95% confidence interval: ±25.789The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:japaneseNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 677 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 628 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlJapaneseReasoning: high | 1,429.848 arena pointsNo clear lead 95% confidence interval: ±25.425The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,418.789 arena points 95% confidence interval: ±26.437The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:japaneseNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 629 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 595 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlKorean | 1,418.422 arena pointsNo clear lead 95% confidence interval: ±19.586The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,393.857 arena points 95% confidence interval: ±21.517The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:koreanNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 1,070 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 869 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlKoreanReasoning: high | 1,431.798 arena pointsWinner 95% confidence interval: ±20.36The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,375.542 arena points 95% confidence interval: ±21.534The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:koreanNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 958 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 919 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlLegal and government | 1,473.155 arena pointsWinner 95% confidence interval: ±9.905The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,448.527 arena points 95% confidence interval: ±10.217The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_legal_and_governmentNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 4,245 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 3,907 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlLegal and governmentReasoning: high | 1,472.404 arena pointsNo clear lead 95% confidence interval: ±10.073The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,458.238 arena points 95% confidence interval: ±10.278The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_legal_and_governmentNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 4,056 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 3,876 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlLife, physical and social science | 1,481.533 arena pointsWinner 95% confidence interval: ±7.371The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,458.241 arena points 95% confidence interval: ±7.633The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_life_and_physical_and_social_scienceNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 8,617 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 7,812 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlLife, physical and social scienceReasoning: high | 1,478.57 arena pointsWinner 95% confidence interval: ±7.454The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,461.582 arena points 95% confidence interval: ±7.609The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_life_and_physical_and_social_scienceNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 8,168 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 7,985 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlLonger queries | 1,472.734 arena pointsWinner 95% confidence interval: ±5.523The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,448.126 arena points 95% confidence interval: ±5.746The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:longer_queryNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 24,005 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 21,643 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlLonger queriesReasoning: high | 1,466.572 arena pointsWinner 95% confidence interval: ±5.598The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,449.175 arena points 95% confidence interval: ±5.755The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:longer_queryNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 22,884 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 21,682 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlMath | 1,444.797 arena pointsNo clear lead 95% confidence interval: ±11.56The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,425.263 arena points 95% confidence interval: ±12.232The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:mathNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 2,862 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,516 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlMathReasoning: high | 1,468.86 arena pointsWinner 95% confidence interval: ±12.286The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,440.663 arena points 95% confidence interval: ±12.368The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:mathNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 2,503 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,448 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlMathematical professions | 1,460.06 arena pointsWinner 95% confidence interval: ±11.553The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,431.855 arena points 95% confidence interval: ±11.996The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_mathematicalNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 3,005 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,738 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlMathematical professionsReasoning: high | 1,468.576 arena pointsWinner 95% confidence interval: ±12.161The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,441.422 arena points 95% confidence interval: ±12.404The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_mathematicalNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 2,683 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 2,588 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlMedicine and healthcare | 1,477.305 arena pointsNo clear lead 95% confidence interval: ±10.595The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,459.52 arena points 95% confidence interval: ±10.979The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_medicine_and_healthcareNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 3,764 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 3,440 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlMedicine and healthcareReasoning: high | 1,485.153 arena pointsWinner 95% confidence interval: ±10.758The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,463.011 arena points 95% confidence interval: ±10.889The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_medicine_and_healthcareNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 3,585 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 3,505 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlMulti-turn | 1,477.081 arena pointsWinner 95% confidence interval: ±7.236The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,452.835 arena points 95% confidence interval: ±7.482The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:multi_turnNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 9,808 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 8,848 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlMulti-turnReasoning: high | 1,455.791 arena pointsNo clear lead 95% confidence interval: ±7.356The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,446.059 arena points 95% confidence interval: ±7.528The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:multi_turnNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 9,151 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 8,708 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlNon-English | 1,445.283 arena pointsWinner 95% confidence interval: ±4.853The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,422.787 arena points 95% confidence interval: ±5.078The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:non_englishNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 30,350 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 27,100 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlNon-EnglishReasoning: high | 1,441.433 arena pointsWinner 95% confidence interval: ±4.93The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,422.674 arena points 95% confidence interval: ±5.146The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:non_englishNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 28,950 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 27,013 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlPolish | 1,462.982 arena pointsNo clear lead 95% confidence interval: ±18.614The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,439.239 arena points 95% confidence interval: ±18.981The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:polishNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 1,067 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,007 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlPolishReasoning: high | 1,462.154 arena pointsNo clear lead 95% confidence interval: ±18.622The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,429.767 arena points 95% confidence interval: ±18.83The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:polishNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 1,049 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,019 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlRussian | 1,458.488 arena pointsWinner 95% confidence interval: ±8.692The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,432.925 arena points 95% confidence interval: ±8.941The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:russianNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 5,725 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 5,272 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlRussianReasoning: high | 1,459.053 arena pointsWinner 95% confidence interval: ±8.778The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,437.146 arena points 95% confidence interval: ±9.061The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:russianNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 5,573 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 5,232 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlSoftware and IT services | 1,492.65 arena pointsWinner 95% confidence interval: ±5.414The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,470.809 arena points 95% confidence interval: ±5.639The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_software_and_it_servicesNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 22,338 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 20,180 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlSoftware and IT servicesReasoning: high | 1,483.295 arena pointsNo clear lead 95% confidence interval: ±5.497The 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 arena points 95% confidence interval: ±5.679The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_software_and_it_servicesNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 21,186 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 20,086 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlSpanish | 1,455.955 arena pointsNo clear lead 95% confidence interval: ±16.149The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,432.449 arena points 95% confidence interval: ±16.767The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:spanishNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 1,656 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,532 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlSpanishReasoning: high | 1,450.859 arena pointsNo clear lead 95% confidence interval: ±16.308The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,432.679 arena points 95% confidence interval: ±17.246The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:spanishNo documented setup differenceNo clear leadDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 1,615 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 1,427 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlWriting, literature and language | 1,451.594 arena pointsWinner 95% confidence interval: ±6.517The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,418.104 arena points 95% confidence interval: ±6.732The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_writing_and_literature_and_languageNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 13,160 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 11,766 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
| Arena Text, Style ControlWriting, literature and languageReasoning: high | 1,443.409 arena pointsWinner 95% confidence interval: ±6.601The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | 1,421.414 arena points 95% confidence interval: ±6.736The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result. | style-control-v1style-controlled Bradley-Terry ratingHigher is betterCohort: lmarena-text-style-control:2026-08-12:industry_writing_and_literature_and_languageNo documented setup differenceWinner: DeepSeek-V4-ProDeepSeek-V4-Pro: Snapshot: deepseek-v4-pro-high-preview, Reasoning: high, Harness: Arena text, style control DeepSeek-V4-Flash: Snapshot: deepseek-v4-flash-high-preview, Reasoning: high, Harness: Arena text, style controlDeepSeek-V4-Pro: Style-controlled Arena rating from 12,742 anonymous pairwise votes.DeepSeek-V4-Flash: Style-controlled Arena rating from 11,864 anonymous pairwise votes. | ArenaType: Official benchmark leaderboardLeaderboard updated: 2026-08-12Import ID: lmarena-eeedf9209373a7f9b324bfb839e7b02fedacd1e5-1611ed64f3f1Retrieved 2026-08-16 |
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.
Prices use official API rates per 1M tokens. Web subscriptions and cloud platform surcharges can differ.
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 DeepSeek-V4-Pro and DeepSeek-V4-Flash with other leading models using the same data and benchmark logic.
View the complete LLM comparison