OpenAI’s GPT-6 Astra recently posted an 82.9% score with tools on the refined HLE-Diamond subset released September 22, 2026, lifting trader expectations for a 55%+ main-exam result by year-end even as Anthropic’s Claude Fable 5.1 and Opus 5.5 hold the top spots on the standard 2,500-question HLE leaderboard at 65% and 61.4%. Full-set OpenAI models such as GPT-5.4 Pro sit around 58%, while verified and rolling variants show Google and Anthropic models trading narrow leads near 54-55%. Key swing factors include OpenAI’s remaining 2026 releases, tool-use protocols, and any late-year leaderboard updates from CAIS or Scale that could shift the highest reported OpenAI accuracy before the December 31 resolution.