Compare / GPT-5.6 Terra vs GPT-6 Sol
GPT-5.6 Terra vs GPT-6 Sol
20 results from 6 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | GPT-5.6 Terra | GPT-6 Sol |
|---|---|---|---|
| APEX-Agentsreported | Tasks passed% of tasks | 58.2% | 54.3% |
| Arena (formerly LMArena)reported | Confirmed task successIPS effect estimate | -0.06 | 0.11 |
| Arena (formerly LMArena)reported | Praise over complaintIPS effect estimate | 0.03 | 0.11 |
| Arena (formerly LMArena)reported | SteerabilityIPS effect estimate | 0.04 | 0.15 |
| Arena (formerly LMArena)reported | Tool groundingIPS effect estimate | 0.00 | 0.00 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,467 | 1,424 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,466 | 1,465 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,428 | 1,451 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,504 | 1,507 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,462 | 1,447 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,465 | 1,457 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,445 | 1,440 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 42.1 | 47.6 |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 42.9% | 47.9% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 83.0% | 83.7% |
| Artificial Analysisreported | SciCode% of problems | 55.0% | 57.6% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 35.4% | 43.9% |
| GDP.pdfreported | Professional document tasks% of rubric | 24.7% | 26.4% |
| SimpleQA Verified (Epoch AI)reported | Correct answers% of questions | 43.2% | 60.7% |
| Vending-Bench 2reported | Money after a yearUS dollars | $7,343.21 | $14,427.85 |
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.