Moonshot AI’s July 2026 release of the open-weights Kimi K3 large language model, scoring 56% on Humanity’s Last Exam with tools and 43.5% without, has anchored trader sentiment around the 50%+ threshold. The 2.8-trillion-parameter model sits just behind leading closed models from Anthropic and OpenAI on current leaderboards, reflecting competitive but not frontier positioning on the 2,500-question expert benchmark. Its native multimodal capabilities, 1M-token context, and open availability support ongoing community fine-tuning and tool integrations that could push scores higher before December 31. Higher thresholds above 60% remain lower probability given benchmark difficulty and the pace of rival releases.