GPT-6.1 Sol vs GPT-6 Sol

OpenAI has officially announced GPT-6.1 Sol, its answer to Claude Opus 5.5. Here is how it compares to GPT-6 Sol:

Benchmarks

BenchmarkGPT-6.1 SolGPT-6 SolDifference
DeepSWE v1.175.2%68.8%GPT-6.1 Sol +6.4 points
GDP.pdf32.0%28.0%GPT-6.1 Sol +4.0 points
AutomationBench 1.0.636.1%33.2%GPT-6.1 Sol +2.9 points
OSWorld 2.0 Offline71.4%64.4%GPT-6.1 Sol +7.0 points
Terminal-Bench Science 0.157.0%27.6%GPT-6.1 Sol +29.4 points
Factual error rate (lower is better)4.1%4.5%GPT-6.1 Sol 0.4 points lower

Pricing and specs

SpecificationGPT-6.1 SolGPT-6 SolDifference
Input, per 1M tokens$2$2Same
Cached input, per 1M tokens$0.10$0.2050% cheaper
Output, per 1M tokens$10$10Same
Context window1.05M tokens1.05M tokensSame
Max output128K tokens128K tokensSame

Artificial Analysis benchmarks

BenchmarkGPT-6.1 Sol (Max)GPT-6 Sol (Max)Difference
Intelligence Index v4.3.251.847.5GPT-6.1 Sol +4.3
AA-Briefcase v1.11564.2 Elo1482.8 EloGPT-6.1 Sol +81.4 Elo
GDPval-AA v2.11575.1 Elo1487 EloGPT-6.1 Sol +88.1 Elo
AutomationBench-AA64.9%62%GPT-6.1 Sol +2.9 points
Terminal-Bench 4.056.1%43.9%GPT-6.1 Sol +12.2 points
SciCode54.2%57.6%GPT-6 Sol +3.4 points
Humanity's Last Exam52.9%48%GPT-6.1 Sol +4.9 points
GDP.pdf31.0%24.8%GPT-6.1 Sol +6.2 points
CritPt31.7%About 31%GPT-6.1 Sol about +0.7 points
AA-Omniscience41.527.1GPT-6.1 Sol +14.4
AA-LCR v1.183.0%83.7%GPT-6 Sol +0.7 points