OpenAI’s GPT-6 Astra currently trails Anthropic’s leading Claude models on standard Humanity’s Last Exam leaderboards, posting roughly 54.7% accuracy versus Claude Fable 5.1 and Opus 5.5 scores of 59–65%. Trader consensus prices 55%+ at high probability because OpenAI maintains a rapid release cadence and has already demonstrated strong tool-augmented results on HLE-Diamond subsets. Key swing factors include any official GPT-6 variants, reasoning optimizations, or verified leaderboard updates before year-end. The benchmark’s expert-curated questions continue to discriminate among frontier large language models, making competitive positioning and verifiable capability jumps the dominant drivers of market-implied odds.