Moonshot AI’s July 2026 launch of the 2.8-trillion-parameter Kimi K3 model, featuring a 1-million-token context window and strong multimodal reasoning, established the current baseline for trader expectations on Humanity’s Last Exam. K3 has posted verified scores in the mid-40s to 56% range across HLE variants, with tools-enabled results near the top of open models but trailing leading closed systems like Claude Fable 5.1 at 65%. Community fine-tuning of its open weights, ongoing inference optimizations, and potential follow-on releases before year-end represent the main upside catalysts. Traders weigh these against the rapid iteration pace of U.S. labs and the benchmark’s resistance to quick saturation, keeping implied probabilities for thresholds above 50% closely contested.