OpenAI’s GPT-6 Astra and related variants currently post mid-50% scores on standard Humanity’s Last Exam (HLE) leaderboards as of mid-October 2026, trailing Anthropic’s Claude Opus 5.5 and Fable 5.1 models that lead several variants at 59–65%. Tool-augmented and HLE-Diamond evaluations show stronger OpenAI results, including an 82.9% mark with web and code tools on a refined 1,000-question subset, highlighting how agentic capabilities and reasoning enhancements can rapidly lift performance. Traders are watching for potential late-2026 OpenAI releases or optimizations that could close the gap before year-end, while noting the benchmark’s resistance to quick saturation and differences across official, verified, and tool-enabled runs. Competitive pressure from Google’s Gemini 4 series and Meta’s Muse models adds uncertainty to whether OpenAI will claim the single highest score by December 31.