Compare / Qwen3 235B A22B Thinking 2507 vs GPT-5.4 Nano

Qwen3 235B A22B Thinking 2507 vs GPT-5.4 Nano

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

Qwen3 235B A22B Thinking 2507
Alibaba · profile
GPT-5.4 Nano
OpenAI · profile
Results compared
19
Shared sources
3
BenchmarkMetricQwen3 235B A22B Thinking 2507GPT-5.4 Nano
Arena (formerly LMArena)reportedBusiness, management and financeArena rating 1,4041,405
Arena (formerly LMArena)reportedCreative writingArena rating 1,3701,335
Arena (formerly LMArena)reportedExpert promptsArena rating 1,4661,438
Arena (formerly LMArena)reportedInstruction followingArena rating 1,3851,387
Arena (formerly LMArena)reportedOverallArena rating 1,4001,401
Arena (formerly LMArena)reportedWriting, literature and languageArena rating 1,3771,362
Artificial AnalysisreportedArtificial Analysis Coding Indexindex score 22.156.1
Artificial AnalysisreportedArtificial Analysis Intelligence Indexindex score 12.720.7
Artificial AnalysisreportedGPQA Diamond% of questions 79.0%81.7%
Artificial AnalysisreportedHumanity's Last Exam% of questions 15.9%28.3%
Artificial AnalysisreportedIFBench% of instructions 51.2%75.9%
Artificial AnalysisreportedLong-context reasoning (AA-LCR)% of questions 72.0%76.7%
Artificial AnalysisreportedSciCode% of problems 41.4%47.2%
Artificial AnalysisreportedTerminal-Bench 2.1% of tasks 12.0%60.7%
Artificial AnalysisreportedTerminal-Bench 4.0% of tasks 0.0%0.5%
Artificial AnalysisreportedTerminal-Bench Hard% of tasks 13.6%42.4%
Artificial AnalysisreportedΤ-bench banking% of tasks 7.8%27.4%
Artificial AnalysisreportedΤ²-bench telecom% of tasks 53.2%76.0%
SimpleQA Verified (Epoch AI)reportedCorrect answers% of questions 40.4%11.7%

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.