OpenAI’s GPT-6 Astra currently posts the strongest reported scores among its lineup on Humanity’s Last Exam variants, reaching 54.7% on Artificial Analysis HLE and 59.9% on the HLE-Diamond subset without tools, with tool-augmented runs climbing to 82.9%. These results place it just behind leading Anthropic models such as Claude Opus 5.5 and Claude Fable 5.1, which top multiple 2026 leaderboards at 59–65%. Rapid iteration across frontier labs, combined with benchmark refinements like HLE-Diamond and verified subsets, has driven double-digit gains this year. With roughly ten weeks left before the December 31 resolution window, any new OpenAI release or post-training update could shift the highest score, though Anthropic’s consistent edge on expert-reasoning tasks continues to pressure the implied probabilities.