OpenAI’s GPT-6 Astra, released in September 2026, currently posts the strongest OpenAI results on Humanity’s Last Exam variants, reaching 54.8% on the full 2,500-question Scale evaluation and 59.9% on the cleaned HLE-Diamond subset without tools. These scores reflect gains from scaled reasoning compute, improved tool use, and post-training refinements that have lifted frontier models roughly 30 points year-over-year. Anthropic’s Claude Opus 5.5 and Fable 5.1 still lead several public leaderboards at 61–65%, creating competitive pressure. Remaining 2026 catalysts include further GPT-6 family iterations, optimized agentic setups, and any HLE-Diamond or Verified updates that could shift reported maxima before year-end resolution.