GPT-6 Astra Sets 98.1 Record on Extended NYT Connections Benchmark

GPT-6 Astra set a new record of 98.1 on the Extended NYT Connections benchmark (xhigh reasoning mode), with the high tier scoring 97.7, while costing roughly 40% less than previous solutions.

2026-09-05 ~ 2026-09-05 · 2 related posts

1 near-duplicate retellings: sherwinwu