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 multimodality, has driven recent gains on Humanity’s Last Exam (HLE), a 2,500-question expert benchmark spanning math, science, and humanities that resists simple retrieval. Kimi K3 has posted around 56% on tool-augmented HLE evaluations and lower without tools, placing it behind leading closed models like Claude Fable 5.1 and Opus 5.5 but competitive with other open systems. Earlier K2 variants reached the mid-40s with agentic tool use. Traders are watching for further Moonshot releases, fine-tuning, or tool improvements before year-end, as HLE’s difficulty makes incremental progress slow and leaderboards update frequently with new configurations.
Polymarketデータを参照したAI生成の実験的な要約。これは取引アドバイスではなく、このマーケットの解決方法には一切関係ありません。 · 更新日View resolved

外部リンクに注意してください。
外部リンクに注意してください。
よくある質問