Kimi

Family

Moonshot AI · Kimi

Moonshot's Kimi family combines multimodal open weights, long-context reasoning, and coding-focused agent models.

chinese long-context reasoning coding agentic multimodal open-weights model-family
Updated July 10, 2026

Overview

This is a model family overview. For version-specific details, see the individual model entries linked below.

Kimi is Moonshot AI’s model family, known for long-context behavior and a practical focus on reasoning, multimodal understanding, coding, and agent workflows. The family has evolved from chat-oriented releases into very large open-weight models with both general-purpose and coding-specialized routes.

Current Latest

Kimi K2.7 Code is the newest coding specialist, released on June 12, 2026 with forced thinking, public weights, multimodal input, and a 262K context. Kimi K2.6 remains Moonshot’s recommended general-purpose route for writing, analysis, conversation, and workflows that need optional non-thinking mode.

Strengths

  • Strong long-context handling for retrieval-heavy and document-centric tasks.
  • Strong coding and reasoning performance in open-weight form.
  • Useful agent and tool-use behavior for multi-step pipelines.
  • Native image and video understanding in current K2.6 and K2.7 routes.
  • Chinese-English bilingual utility for mixed-language products.
  • Flexible deployment via open weights or managed services.

When to Choose Kimi

Choose Kimi when you need open-weight control with strong long-context reasoning and coding quality. Use K2.7 Code for thinking-heavy engineering agents and K2.6 for broader multimodal assistant work. Both are very large models that usually require managed or serious private infrastructure.

Access

  • Moonshot AI/Kimi APIs, Kimi Code, and product surfaces
  • Official open-weight releases on Hugging Face
  • Community and partner inference stacks for self-hosting