Compare / Claude Opus 5 vs GPT-6 Sol
Claude Opus 5 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 | Claude Opus 5 | GPT-6 Sol |
|---|---|---|---|
| APEX-Agentsreported | Tasks passed% of tasks | 65.8% | 54.3% |
| Arena (formerly LMArena)reported | Confirmed task successIPS effect estimate | 0.10 | 0.11 |
| Arena (formerly LMArena)reported | Praise over complaintIPS effect estimate | 0.18 | 0.11 |
| Arena (formerly LMArena)reported | SteerabilityIPS effect estimate | 0.11 | 0.15 |
| Arena (formerly LMArena)reported | Tool groundingIPS effect estimate | 0.00 | 0.00 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,488 | 1,424 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,486 | 1,465 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,472 | 1,451 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,543 | 1,507 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,497 | 1,447 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,491 | 1,457 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,482 | 1,440 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 50.8 | 47.6 |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 54.9% | 47.9% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 82.0% | 83.7% |
| Artificial Analysisreported | SciCode% of problems | 56.4% | 57.6% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 49.0% | 43.9% |
| GDP.pdfreported | Professional document tasks% of rubric | 24.0% | 26.4% |
| SimpleQA Verified (Epoch AI)reported | Correct answers% of questions | 59.9% | 60.7% |
| Vending-Bench 2reported | Money after a yearUS dollars | $11,181.87 | $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.