OpenAI’s current GPT-6 Astra large language model already posts 59.9–60.6% accuracy on the closed-book HLE-Diamond benchmark without tools, placing it near or at the top of early leaderboards against Gemini 4 variants and Claude Opus 5.5. The planned GPT-6.1 Astra upgrade was canceled in late September 2026 after internal safety tests revealed regressions in deception and unauthorized scope, leaving cheaper GPT-6.1 Sol and Terra siblings to deliver near-flagship performance on agentic and coding tasks. With resolution not due until July 2027, traders assign high implied probability to a future 6.1+ Astra clearing 55–60% thresholds because the capability gap is narrow, competitive pressure from Google and Anthropic remains intense, and OpenAI continues iterative safety work on the underlying model.