Kimi K2.7 Code

K2.7 Code · Family: Kimi

Coding-focused agentic MoE model (1T params, 32B active) from Moonshot AI, built on Kimi K2.6; thinking mode, 256K context, text/image/video input.

✓ Active✓ Public access⚖ Open weightsLLMMultimodalReasoning modelTool-using model📁 Kimi

Context window

256K

tokens

Parameters

1T (32B aktywnych)

parameters

Release date

11 June 2026

🏢Moonshot AIProducer

Access:APIDownloadHostedDeployment:💻 Local☁ Cloud

Overview

Kimi K2.7 Code is a coding-focused agentic language model developed by Moonshot AI and built on top of Kimi K2.6. It uses a Mixture-of-Experts (MoE) architecture with 1 trillion total parameters, of which 32 billion are activated per token.

Architecture

The model has 61 layers (including 1 dense layer), 384 experts with 8 selected per token and 1 shared expert. It uses the MLA attention mechanism, the SwiGLU activation function and the MoonViT vision encoder (400M parameters). The context window is 256K tokens and the vocabulary size is 160K.

Behavior and availability

The model runs in thinking mode by default and preserves reasoning content across turns (preserve_thinking). It accepts text, image and video input, supports tool calling (including the MCP protocol) and native INT4 quantization. Compared with Kimi K2.6 it reduces thinking-token usage by roughly 30%. Weights are released under a Modified MIT license; the model is available via API (Moonshot AI platform, OpenAI/Anthropic-compatible), as a download on Hugging Face and through the Kimi Code service.

Classification

LLMMultimodalReasoning modelTool-using model

Family: Kimi

Access & deployment

APIDownloadHosted

LocalCloud

Weights: Open weights

Key parameters

📏 Context: 256K

🧩 Parameters: 1T (32B aktywnych)

✓ Tools

📥 Input: text, image, video

Technical specification

Context window

256K

tokens

Parameters

1T (32B aktywnych)

parameters

License

Modified MIT

Features:✓ Tool use

Modalities

⬇ Input

textimagevideo

⬆ Output

textcode

Capabilities and applications

Native model capabilities

Coding

Generating, analysing and modifying code in many programming languages. Covers writing functions, debugging, refactoring, code review, and creating tests. Measured by benchmarks such as HumanEval and SWE-bench.

Category: coding

Agentic capability

The model's ability to autonomously plan and execute multi-step tasks by sequentially using tools, maintaining context, and adapting to intermediate results.

Category: planning

Reasoning

The model's ability to reason logically and solve complex problems.

Category: reasoning

Multi-step reasoning

Carrying out multi-step chains of reasoning across long, complex tasks.

Category: reasoning

Long context

Support for large context windows — tens to hundreds of thousands (or millions) of input tokens. Enables analysis of entire codebases, long documents, and many parallel conversations without losing earlier information. GPT-5.1 supports 400,000 tokens.

Category: language

Image understanding

Analysing and interpreting the content of images.

Category: vision

Video understanding

The model's ability to analyse and interpret video content — recognising actions, motion, events and relationships between objects over time.

Category: video

Parallel Tool Calls

Ability to invoke multiple external tools simultaneously while generating a response.

Category: reasoning

Planning

Forming and executing action plans for complex tasks.

Category: planning

Multimodal understanding

Category: multimodal

Structured output

Producing data in structured formats such as JSON.

Category: structured_generation

Benchmark results

6 benchmarks

Kimi Code Bench v2

thinking mode, Kimi Code CLI, 262,144-token context, temp=1.0, top-p=0.95

62.0