Compare / Mistral Medium 3.5 vs GLM 5.2

Mistral Medium 3.5 vs GLM 5.2

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

Mistral Medium 3.5
Mistral · profile
GLM 5.2
Z.ai · profile
Results compared
26
Shared sources
3
BenchmarkMetricMistral Medium 3.5GLM 5.2
Arena (formerly LMArena)reportedConfirmed task successIPS effect estimate -0.160.04
Arena (formerly LMArena)reportedPraise over complaintIPS effect estimate -0.150.10
Arena (formerly LMArena)reportedSteerabilityIPS effect estimate -0.070.04
Arena (formerly LMArena)reportedTool groundingIPS effect estimate -0.020.00
Arena (formerly LMArena)reportedOverallArena rating 1,4301,463
Arena (formerly LMArena)reportedBusiness, management and financeArena rating 1,4301,463
Arena (formerly LMArena)reportedCreative writingArena rating 1,3981,455
Arena (formerly LMArena)reportedExpert promptsArena rating 1,4341,500
Arena (formerly LMArena)reportedInstruction followingArena rating 1,4211,470
Arena (formerly LMArena)reportedOverallArena rating 1,4261,476
Arena (formerly LMArena)reportedWriting, literature and languageArena rating 1,4041,459
Artificial AnalysisreportedArtificial Analysis Coding Indexindex score 46.968.8
Artificial AnalysisreportedArtificial Analysis Intelligence Indexindex score 14.233.7
Artificial AnalysisreportedGPQA Diamond% of questions 74.8%89.5%
Artificial AnalysisreportedHumanity's Last Exam% of questions 13.8%41.1%
Artificial AnalysisreportedIFBench% of instructions 68.8%73.3%
Artificial AnalysisreportedLong-context reasoning (AA-LCR)% of questions 69.3%78.3%
Artificial AnalysisreportedSciCode% of problems 40.2%51.2%
Artificial AnalysisreportedTerminal-Bench 2.1% of tasks 50.6%77.9%
Artificial AnalysisreportedTerminal-Bench 4.0% of tasks 0.0%1.0%
Artificial AnalysisreportedTerminal-Bench Hard% of tasks 33.3%50.8%
Artificial AnalysisreportedΤ-bench banking% of tasks 15.1%34.6%
Artificial AnalysisreportedΤ²-bench telecom% of tasks 94.2%99.1%
UGI LeaderboardreportedRequested-length error% off the requested word count 18.0%12.0%
UGI LeaderboardreportedStyle adherencescore from 0 to 1 0.350.37
UGI LeaderboardreportedWriting scorescore out of 100 45.567.0

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.