概述
Its releases range from long-context APIs to reasoning and mixture-of-experts models, so context limits and capabilities should be tied to exact model versions.
深入探討
Moonshot AI’s official timeline dates its founding to Spring 2023. The company opened the moonshot-v1 API beta on January 31, 2024, with up to 128K context. It later released Kimi K1.5, a reasoning model described in its research paper, and Kimi K2, whose technical report describes a mixture-of-experts architecture with 1 trillion total parameters and 32 billion activated. Moonshot’s product and API model limits change across releases; confirm the selected model and input budget before submitting long documents. Alibaba’s FY2024 annual report disclosed an investment in Moonshot AI, including an approximately 36% equity interest during that fiscal year.
現實世界的實施
上傳 200 頁的法律合約並要求 Kimi 總結義務並標記異常條款
貼上整篇學術論文或多篇論文以獲得文獻綜述式的綜合
向 Kimi K2 提供大型程式碼庫以定位錯誤並解釋模組如何交互
分析公司冗長的年度報告以提取收入趨勢和風險因素
Kimi 和 Moonshot AI 的未來
Moonshot publishes model updates and research that change context limits, agent capabilities, and access. Compare exact versions and read current model documentation before relying on a capability, benchmark, or input limit; API and consumer products may not expose identical models.
付諸實踐
為什麼這很重要
- 供應商策略. 供應商路線圖會影響您的團隊接下來可以建立的功能。
- 成本與預算. 商業條款和部署選項會影響長期成本和風險。
- 風險與安全. 公司激勵措施塑造了產品預設、安全態勢和開放性。
留意
- 發佈公告可能會超過實際生產工作流程的穩定性。
- API 定價或政策轉變可能會在一夜之間打破假設。
- 單一供應商依賴性增加了鎖定和遷移成本。
後續步驟
- 使用您自己的任務和資料集評估提供者。
- 在整合之前查看隱私、安全和法律條款。
- 維護跨模型或供應商的後備計劃。
- 監控發行說明,以便路線圖的變更不會讓團隊感到意外。
不斷探索
常見問題
Kimi 和 Moonshot AI 是什麼?
Moonshot AI develops the Kimi model family and assistant. Its releases range from long-context APIs to reasoning and mixture-of-experts models, so context limits and capabilities should be tied to exact model versions.
Moonshot 在 2024 年 1 月公開測試版中宣布了 Moonshot-v1 API 的最大上下文大小是多少?
Moonshot’s official Kimi Latest article says the API public beta launched January 31, 2024 with moonshot-v1 models supporting up to 128K context.
阿里巴巴在2024財年年報中揭露了哪些主要投資者?
Alibaba’s annual report states that it invested about US$0.8 billion for an approximately 36% equity interest in Moonshot during FY2024.
後來的 Kimi 模型(例如 K2)使用什麼架構來維持大規模的效率?
Kimi K2 採用專家混合設計,每個令牌僅啟動一部分參數。
2025 年初發布的 Kimi k1.5 旨在改進什麼?
Kimi k1.5 是一個以推理為中心的模型,旨在增強多步驟問題解決能力。
為什麼長上下文視窗實際上很有用?
長上下文視窗意味著模型可以在單一查詢中處理整本書或大型程式碼庫。