GPT-6 Astra has achieved a new top score of 98.1 on the Extended NYT Connections Benchmark, surpassing previous models. A variant, GPT-6 Astra (xhigh), also scored exceptionally high at 97.7. Both GPT-6 Astra models demonstrated superior performance compared to GPT-5.6 Sol, while operating at approximately 40% lower cost per puzzle. AI
IMPACT Sets a new performance standard on a popular benchmark, potentially influencing future model development and cost-efficiency targets.
RANK_REASON New benchmark score for a model. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →