Compare / Claude Sonnet 4.6 vs DeepSeek V4 Flash 0423

Claude Sonnet 4.6 vs DeepSeek V4 Flash 0423

19 results from 3 sources that measured both models. Each row is in the source's own units; there is no overall winner.

Claude Sonnet 4.6
Anthropic · profile
DeepSeek V4 Flash 0423
DeepSeek · profile
Results compared
19
Shared sources
3
BenchmarkMetricClaude Sonnet 4.6DeepSeek V4 Flash 0423
Arena (formerly LMArena)reportedOverallArena rating 1,4611,439
Arena (formerly LMArena)reportedBusiness, management and financeArena rating 1,4821,439
Arena (formerly LMArena)reportedCreative writingArena rating 1,4491,408
Arena (formerly LMArena)reportedExpert promptsArena rating 1,5061,457
Arena (formerly LMArena)reportedInstruction followingArena rating 1,4751,428
Arena (formerly LMArena)reportedOverallArena rating 1,4721,436
Arena (formerly LMArena)reportedWriting, literature and languageArena rating 1,4551,416
Artificial AnalysisreportedArtificial Analysis Coding Indexindex score 63.069.1
Artificial AnalysisreportedArtificial Analysis Intelligence Indexindex score 30.134.3
Artificial AnalysisreportedGPQA Diamond% of questions 87.5%90.8%
Artificial AnalysisreportedHumanity's Last Exam% of questions 33.6%38.6%
Artificial AnalysisreportedLong-context reasoning (AA-LCR)% of questions 80.0%79.7%
Artificial AnalysisreportedSciCode% of problems 50.1%50.3%
Artificial AnalysisreportedTerminal-Bench 2.1% of tasks 71.2%78.7%
Artificial AnalysisreportedTerminal-Bench 4.0% of tasks 3.0%12.1%
Artificial AnalysisreportedΤ-bench banking% of tasks 34.4%39.4%
UGI LeaderboardreportedRequested-length error% off the requested word count 2.0%12.0%
UGI LeaderboardreportedStyle adherencescore from 0 to 1 0.410.36
UGI LeaderboardreportedWriting scorescore out of 100 64.554.6

Bold green marks the better value on that metric. Where a source reports ranges that overlap, the difference may not be meaningful; see the benchmark page for ranges.