Compare / GPT-5.5 vs GLM 5.3

GPT-5.5 vs GLM 5.3

23 results from 5 sources that measured both models. Each row is in the source's own units; there is no overall winner.

GPT-5.5
OpenAI · profile
GLM 5.3
Z.ai · profile
Results compared
23
Shared sources
5
BenchmarkMetricGPT-5.5GLM 5.3
APEX-AgentsreportedTasks passed% of tasks 55.1%56.6%
Arena (formerly LMArena)reportedConfirmed task successIPS effect estimate -0.030.06
Arena (formerly LMArena)reportedPraise over complaintIPS effect estimate 0.080.06
Arena (formerly LMArena)reportedSteerabilityIPS effect estimate 0.070.03
Arena (formerly LMArena)reportedTool groundingIPS effect estimate 0.000.00
Arena (formerly LMArena)reportedOverallArena rating 1,4861,468
Arena (formerly LMArena)reportedBusiness, management and financeArena rating 1,4891,474
Arena (formerly LMArena)reportedCreative writingArena rating 1,4521,456
Arena (formerly LMArena)reportedExpert promptsArena rating 1,5131,516
Arena (formerly LMArena)reportedInstruction followingArena rating 1,4781,480
Arena (formerly LMArena)reportedOverallArena rating 1,4811,480
Arena (formerly LMArena)reportedWriting, literature and languageArena rating 1,4701,470
Artificial AnalysisreportedArtificial Analysis Coding Indexindex score 74.974.8
Artificial AnalysisreportedArtificial Analysis Intelligence Indexindex score 38.444.8
Artificial AnalysisreportedGPQA Diamond% of questions 93.5%91.7%
Artificial AnalysisreportedHumanity's Last Exam% of questions 45.8%42.3%
Artificial AnalysisreportedLong-context reasoning (AA-LCR)% of questions 84.3%79.7%
Artificial AnalysisreportedSciCode% of problems 56.1%59.0%
Artificial AnalysisreportedTerminal-Bench 2.1% of tasks 84.3%83.9%
Artificial AnalysisreportedTerminal-Bench 4.0% of tasks 14.6%41.9%
Artificial AnalysisreportedΤ-bench banking% of tasks 39.0%50.3%
SimpleQA Verified (Epoch AI)reportedCorrect answers% of questions 63.0%41.0%
Vending-Bench 2reportedMoney after a yearUS dollars $7,523.84$8,163.61

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.