Compare / GPT-5.4 Nano vs GPT-5.6 Terra
GPT-5.4 Nano vs GPT-5.6 Terra
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 Nano | GPT-5.6 Terra |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,418 | 1,467 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,405 | 1,466 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,335 | 1,428 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,438 | 1,504 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,387 | 1,462 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,401 | 1,465 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,362 | 1,445 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 56.1 | 76.7 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 20.7 | 42.1 |
| Artificial Analysisreported | GPQA Diamond% of questions | 81.7% | 92.5% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 28.3% | 42.9% |
| Artificial Analysisreported | IFBench% of instructions | 75.9% | 71.2% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 76.7% | 83.0% |
| Artificial Analysisreported | SciCode% of problems | 47.2% | 55.0% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 60.7% | 88.0% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 0.5% | 35.4% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 42.4% | 62.9% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 27.4% | 40.2% |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 76.0% | 86.3% |
| SimpleQA Verified (Epoch AI)reported | Correct answers% of questions | 11.7% | 43.2% |
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.