Kimi K2.7 Code

• 🧠 Coding-focused agentic model built on Kimi K2.6.
• 🏢 Made by Moonshot AI, released June 2026.
• 📏 1T total parameters, 32B active, 256K context.
• 🔧 Mixture-of-Experts; always operates in thinking mode.
• 👁️ Accepts text and image input for coding workflows.
• 🔧 Supports function-calling and web-search for agentic loops.
• 🎯 Targets long-horizon software engineering and tool use.
• 🔒 Open weights published on Hugging Face.

Moonshot

Moonshot is an AI research lab known for developing the Kimi family of large language models. The organization has gained recognition for building capable reasoning-oriented models, with the Kimi line representing its flagship series of text generation…

Kimi K2.7 Code is Moonshot AI's coding-specialized member of the Kimi K2 line, trained directly on top of the general-purpose Kimi K2.6 released two months earlier, which itself succeeded Kimi K2.5. Rather than a broad capability bump, it is a focused agentic-coding release that keeps the trillion-parameter Mixture-of-Experts architecture (1T total, 32B active per token) and adds long-horizon software-engineering training for tasks like codebase analysis, debugging, refactoring, and multi-step tool use.

Architecturally it stays close to its predecessors, so existing deployment setups can largely be reused, and it ships with a 256K-token context window and always-on thinking mode. The Venice catalog lists it as supporting text and image input with function-calling and web-search capabilities, distributed here at int4 quantization.

Compared with its predecessor, Moonshot reports gains on its own coding and agent evaluations, attributing the improvement to the model's tool-calling and long-horizon software-engineering focus; these figures are vendor self-reported on internal benchmarks, so treat them as the provider's claims rather than independent results, and third-party evaluation data was limited at release.

The weights are published on Hugging Face, and the model is paired with Moonshot's coding agent tooling for terminal and multi-turn workflows.