Compare / GPT-5.1 vs GPT-5.4 Nano
GPT-5.1 vs GPT-5.4 Nano
20 results from 4 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | GPT-5.1 | GPT-5.4 Nano |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,459 | 1,418 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,448 | 1,405 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,428 | 1,335 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,482 | 1,438 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,450 | 1,387 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,456 | 1,401 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,437 | 1,362 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 49.4 | 56.1 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 24.7 | 20.7 |
| Artificial Analysisreported | GPQA Diamond% of questions | 87.3% | 81.7% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 28.5% | 28.3% |
| Artificial Analysisreported | IFBench% of instructions | 72.9% | 75.9% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 80.0% | 76.7% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 52.4% | 60.7% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 45.5% | 42.4% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 15.9% | 27.4% |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 81.9% | 76.0% |
| SimpleQA Verified (Epoch AI)reported | Correct answers% of questions | 48.0% | 11.7% |
| Vectara Hallucination Leaderboardreported | Answer rate% of documents | 100.0% | 100.0% |
| Vectara Hallucination Leaderboardreported | Hallucination rate% of summaries | 10.9% | 3.1% |
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.