Compare / GPT-5.4 Mini vs Grok 4.3
GPT-5.4 Mini vs Grok 4.3
20 results from 3 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | GPT-5.4 Mini | Grok 4.3 |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,447 | 1,428 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,460 | 1,440 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,402 | 1,423 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,480 | 1,452 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,434 | 1,415 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,447 | 1,442 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,424 | 1,424 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 56.1 | 42.2 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 24.1 | 24.9 |
| Artificial Analysisreported | GPQA Diamond% of questions | 87.5% | 90.1% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 28.1% | 37.2% |
| Artificial Analysisreported | IFBench% of instructions | 73.3% | 83.3% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 77.0% | 75.0% |
| Artificial Analysisreported | SciCode% of problems | 52.1% | 48.3% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 59.2% | 39.7% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 2.0% | 0.0% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 52.3% | 37.9% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 25.6% | 12.4% |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 83.3% | 97.7% |
| SimpleQA Verified (Epoch AI)reported | Correct answers% of questions | 29.4% | 33.2% |
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.