Benchmarks
| Benchmark | GPT-6.1 Sol | Claude Opus 5.5 | Difference |
|---|
| GDP.pdf | 32.0%High effort | 28.8%High effort | GPT-6.1 Sol +3.2 points |
| AutomationBench 1.0.6 | 36.1%Max effort | 42.5%Max effort | Opus +6.4 points |
| AutomationBench 1.0.6 | 31.7%Medium effort | 29.5%Medium effort | GPT-6.1 Sol +2.2 points |
| Terminal-Bench Science 0.1 | 57.0%Max effort | 63.3%Max effort | Opus +6.3 points |
Artificial Analysis benchmarks
| Benchmark | GPT-6.1 Sol (Max) | Claude Opus 5.5 (Max) | Difference |
|---|
| Intelligence Index v4.3.2 | 51.8 | 57.6 | Opus +5.8 |
| AA-Briefcase v1.1 | 1564.2 Elo | 1822 Elo | Opus +257.8 Elo |
| GDPval-AA v2.1 | 1575.1 Elo | 1846 Elo | Opus +270.9 Elo |
| AutomationBench-AA | 64.9% | 70% | Opus +5.1 points |
| Terminal-Bench 4.0 | 56.1% | 60% | Opus +3.9 points |
| SciCode | 54.2% | 67% | Opus +12.8 points |
| Humanity's Last Exam | 52.9% | 61% | Opus +8.1 points |
| GDP.pdf | 31.0% | 26% | GPT-6.1 Sol +5.0 points |
| CritPt | 31.7% | 32% | Opus +0.3 points |
| AA-Omniscience | 41.5 | 46 | Opus +4.5 |
| AA-LCR v1.1 | 83.0% | 85% | Opus +2.0 points |
Cost and efficiency
| Metric | GPT-6.1 Sol (Max) | Claude Opus 5.5 (Max) | Difference |
|---|
| Intelligence Index | 51.8 | 57.6 | Opus +5.8 |
| Cost per Intelligence Index task | $0.72 | $5.98 | Sol about 88% cheaper |
| Output tokens across the index | About 67M | About 260M | Sol uses about 74% fewer |
| Input, per 1M tokens | $2 | $4 | Sol 50% cheaper |
| Output, per 1M tokens | $10 | $20 | Sol 50% cheaper |
| Cached input, per 1M tokens | $0.10 | $0.20 | Sol 50% cheaper |
Pricing and specs
| Specification | GPT-6.1 Sol | Claude Opus 5.5 | Difference |
|---|
| Input, per 1M tokens | $2 | $4 | Sol 50% cheaper |
| Cached input, per 1M tokens | $0.10 | $0.20 | Sol 50% cheaper |
| Output, per 1M tokens | $10 | $20 | Sol 50% cheaper |
| Context window | 1.05M tokens | 1M tokens | Sol +50K tokens |
| Max output | 128K tokens | 128K tokens | Same |
| Reasoning levels | Low to Max | Adaptive reasoning, effort levels | Both adjustable |
