Benchmarks
| Benchmark | GPT-6.1 Sol | GPT-6 Astra | Difference |
|---|
| DeepSWE v1.1 | 75.2% | 74.1% | GPT-6.1 Sol +1.1 points |
| GDP.pdf | 32.0% | 32.2% | Astra +0.2 points |
| AutomationBench 1.0.6 | 36.1% | 41.4% | Astra +5.3 points |
| OSWorld 2.0 Offline | 71.4% | 73.5% | Astra +2.1 points |
| Terminal-Bench Science 0.1 | 57.0% | 68.1% | Astra +11.1 points |
| Factual error rate, best (lower is better) | 4.1% | 3.9% | Astra 0.2 points lower |
Artificial Analysis benchmarks
| Benchmark | GPT-6.1 Sol (Max) | GPT-6 Astra (Max) | Difference |
|---|
| Intelligence Index v4.3.2 | 51.8 | 52.7 | Astra +0.9 |
| AA-Briefcase v1.1 | 1564.2 Elo | 1568.9 Elo | Astra +4.7 Elo |
| GDPval-AA v2.1 | 1575.1 Elo | 1541.9 Elo | GPT-6.1 Sol +33.2 Elo |
| AutomationBench-AA | 64.9% | 68.5% | Astra +3.6 points |
| Terminal-Bench 4.0 | 56.1% | 59.1% | Astra +3.0 points |
| SciCode | 54.2% | 56.5% | Astra +2.3 points |
| Humanity's Last Exam | 52.9% | 54.7% | Astra +1.8 points |
| GDP.pdf | 31.0% | 31.0% | Tie |
| CritPt | 31.7% | 31.7% | Tie |
| AA-Omniscience | 41.5 | 43.4 | Astra +1.9 |
| AA-LCR v1.1 | 83.0% | 80.7% | GPT-6.1 Sol +2.3 points |
Cost and efficiency
| Metric | GPT-6.1 Sol (Max) | GPT-6 Astra (Max) | Difference |
|---|
| Intelligence Index | 51.8 | 52.7 | Astra +0.9 |
| Cost per Intelligence Index task | $0.72 | $3.26 | Sol about 78% cheaper |
| Input, per 1M tokens | $2 | $10 | Sol 80% cheaper |
| Cached input, per 1M tokens | $0.10 | $1.00 | Sol 90% cheaper |
| Output, per 1M tokens | $10 | $50 | Sol 80% cheaper |
Pricing and specs
| Specification | GPT-6.1 Sol | GPT-6 Astra | Difference |
|---|
| Input, per 1M tokens | $2 | $10 | Sol 5x cheaper |
| Cached input, per 1M tokens | $0.10 | $1.00 | Sol 10x cheaper |
| Output, per 1M tokens | $10 | $50 | Sol 5x cheaper |
| Context window | 1.05M tokens | 1.05M tokens | Same |
| Max output | 128K tokens | 128K tokens | Same |
| Knowledge cutoff | April 30, 2026 | April 30, 2026 | Same |
| Reasoning effort | Low to Max | Low to Max | Same |
| Text and image input | Yes | Yes | Same |
