Compare / DeepSeek V4 Pro 0423 vs Gemini 3.8 Flash

DeepSeek V4 Pro 0423 vs Gemini 3.8 Flash

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

DeepSeek V4 Pro 0423
DeepSeek · profile
Gemini 3.8 Flash
Google · profile
Results compared
23
Shared sources
3
BenchmarkMetricDeepSeek V4 Pro 0423Gemini 3.8 Flash
ROASBenchmeasured by usAudience scorescore out of 100 44.867.1
ROASBenchmeasured by usBusiness scorescore out of 100 23.851.0
ROASBenchmeasured by usConsistency scorescore out of 100 9.136.4
ROASBenchmeasured by usContribution profitsimulated £ 301,289842,432
ROASBenchmeasured by usCost of a runUS dollars $0.20$0.16
ROASBenchmeasured by usMonths over budgetmonths of 12 00
ROASBenchmeasured by usOverall scorescore out of 100 28.251.4
ROASBenchmeasured by usPlanning scorescore out of 100 54.054.7
ROASBenchmeasured by usReturn on ad spendprofit per £1 spent £0.26£0.65
Arena (formerly LMArena)reportedConfirmed task successIPS effect estimate -0.040.09
Arena (formerly LMArena)reportedPraise over complaintIPS effect estimate -0.020.05
Arena (formerly LMArena)reportedSteerabilityIPS effect estimate -0.030.04
Arena (formerly LMArena)reportedTool groundingIPS effect estimate -0.000.00
Arena (formerly LMArena)reportedOverallArena rating 1,4551,477
Arena (formerly LMArena)reportedBusiness, management and financeArena rating 1,4601,482
Arena (formerly LMArena)reportedCreative writingArena rating 1,4461,482
Arena (formerly LMArena)reportedExpert promptsArena rating 1,4801,529
Arena (formerly LMArena)reportedInstruction followingArena rating 1,4521,488
Arena (formerly LMArena)reportedOverallArena rating 1,4581,492
Arena (formerly LMArena)reportedWriting, literature and languageArena rating 1,4511,486
UGI LeaderboardreportedRequested-length error% off the requested word count 34.0%13.0%
UGI LeaderboardreportedStyle adherencescore from 0 to 1 0.360.34
UGI LeaderboardreportedWriting scorescore out of 100 68.478.5

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.