Compare / DeepSeek V3.1 Terminus vs Gemma 4 26B A4B

DeepSeek V3.1 Terminus vs Gemma 4 26B A4B

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

DeepSeek V3.1 Terminus
DeepSeek · profile
Gemma 4 26B A4B
Google · profile
Results compared
19
Shared sources
3
BenchmarkMetricDeepSeek V3.1 TerminusGemma 4 26B A4B
Arena (formerly LMArena)reportedBusiness, management and financeArena rating 1,4231,444
Arena (formerly LMArena)reportedCreative writingArena rating 1,4051,399
Arena (formerly LMArena)reportedInstruction followingArena rating 1,4201,437
Arena (formerly LMArena)reportedOverallArena rating 1,4171,438
Arena (formerly LMArena)reportedWriting, literature and languageArena rating 1,4111,406
Artificial AnalysisreportedArtificial Analysis Coding Indexindex score 43.539.3
Artificial AnalysisreportedArtificial Analysis Intelligence Indexindex score 14.816.7
Artificial AnalysisreportedGPQA Diamond% of questions 79.2%79.2%
Artificial AnalysisreportedHumanity's Last Exam% of questions 16.4%19.3%
Artificial AnalysisreportedIFBench% of instructions 57.0%72.4%
Artificial AnalysisreportedLong-context reasoning (AA-LCR)% of questions 69.3%65.7%
Artificial AnalysisreportedSciCode% of problems 38.0%40.0%
Artificial AnalysisreportedTerminal-Bench 2.1% of tasks 44.9%39.0%
Artificial AnalysisreportedTerminal-Bench Hard% of tasks 31.8%25.0%
Artificial AnalysisreportedΤ-bench banking% of tasks 21.0%12.0%
Artificial AnalysisreportedΤ²-bench telecom% of tasks 37.1%43.6%
UGI LeaderboardreportedRequested-length error% off the requested word count 32.0%8.0%
UGI LeaderboardreportedStyle adherencescore from 0 to 1 0.370.35
UGI LeaderboardreportedWriting scorescore out of 100 54.043.8

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.