Anthropic’s Claude Opus 5.5 currently leads verified HLE leaderboards at 61.4% (standard) to 67.7% (with tools) as of October 10, 2026, ahead of its own Fable 5.1 and OpenAI’s GPT-6 Astra variants. Rapid capability gains—roughly 30 points in the past year—stem from larger models, improved reasoning chains, and tool integration, with Anthropic’s September releases driving the latest jumps. OpenAI’s Astra demonstrated 82.9% on a refined HLE-Diamond subset, signaling that agentic or high-compute setups could push frontier scores higher before year-end. Key swing factors include December model drops from Google, Meta, or xAI, plus any official HLE updates expanding the question set or evaluation rules. Traders price in continued scaling while noting benchmark saturation risks and lab-specific access differences.