Anthropic’s Claude Fable 5.1 and related Opus/Mythos variants currently lead Humanity’s Last Exam leaderboards at 59–65% as of early September 2026, driven by iterative reasoning improvements, extended context handling, and tool-augmented evaluation on the 2,500-question expert-curated benchmark spanning mathematics, physics, and other domains. This rapid climb from mid-40s percent earlier in the year reflects competitive pressure among frontier labs, with OpenAI’s GPT-5.4/GPT-6 Astra and Meta’s Muse Spark 1.1 trailing by several points while Google and smaller players post lower figures. Trader sentiment on the overall 2026 high centers on continued model cadence and reasoning optimizations before year-end, tempered by the benchmark’s design to resist quick saturation and the possibility of evaluation-condition differences across sources. Key near-term catalysts include additional 2026 releases and any official tool-use or multimodal updates that could shift the frontier.