Moonshot AI’s July 2026 release of Kimi K3, a 2.8-trillion-parameter open-weights model with a 1-million-token context window and native vision, drives current trader sentiment on the Polymarket. It has posted scores in the mid-to-high 40s percent on Humanity’s Last Exam without tools and up to 56 percent with tool use, placing it behind Anthropic’s leading Claude Fable 5.1 and Opus 5 variants on multiple leaderboards. Ongoing fine-tuning, agentic workflows, and community contributions enabled by the open release could push scores higher before year-end, while competitive pressure from GPT-5 series and Gemini updates adds uncertainty. Key catalysts include additional optimization releases, expanded tool integration, and any new Moonshot models that could shift the highest achievable Kimi score on the expert-level benchmark.