# Kimi K2-Instruct-0905

> Moonshot AI's trillion-parameter open-weight MoE, tuned for agentic coding

- Maker: Moonshot AI (Kimi)
- Line: Kimi K2
- Canonical page: https://ai-tldr.dev/models/kimi-k2-instruct-0905/
- Updated: 2026-06-18

## In short

Kimi K2-Instruct-0905 is an updated checkpoint of Moonshot AI's Kimi K2 model line, released on September 9, 2025. It has a 256K-token context window and costs $0.60 per million input tokens and $2.50 per million output tokens. You can use it through Moonshot AI (Kimi) API, OpenRouter and Groq.

## Facts

| Fact | Value | Source |
| --- | --- | --- |
| Context window | 256K (max output ~33K tokens) | [1] |
| Input price | $0.60 / 1M tokens | [2] |
| Output price | $2.50 / 1M tokens | [2] |
| Licence | Modified MIT License · open weights | [1] |
| Released | 9 September 2025 | [3] |
| Parameters | 1T total / 32B active (MoE) | [1] |
| Knowledge cutoff | Not publicly disclosed | [1] |

## Sources

1. [Hugging Face model card](https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905)
2. [OpenRouter pricing & benchmarks](https://openrouter.ai/moonshotai/kimi-k2-0905)
3. [Kimi K2 GitHub repository](https://github.com/MoonshotAI/Kimi-K2)

## Overview

Kimi K2-Instruct-0905 is an updated checkpoint of Moonshot AI's Kimi K2 model line, released on September 9, 2025. Like the original Kimi K2-Instruct, it is a Mixture-of-Experts (MoE) language model with 1 trillion total parameters but only 32 billion activated per token, which keeps inference costs closer to a mid-sized dense model while drawing on the capacity of a much larger one. The weights are published openly under a Modified MIT License on Hugging Face.

The 0905 update focuses on three things: stronger agentic coding (higher accuracy and better generalization across different coding scaffolds and harnesses), a more polished frontend-coding experience (more aesthetic and functional output for web, 3D, and UI tasks), and an extended context window, raised from 128K to 256K tokens to better handle multi-file edits, large-repository question answering, and long-horizon agent runs.

Architecturally it uses 384 experts with 8 routed plus 1 shared expert selected per token, 61 layers, 64 attention heads with MLA attention, and SwiGLU activations. It is a text-only, non-reasoning instruct model, positioned for tool use and coding agents rather than long chain-of-thought reasoning. It can be self-hosted on vLLM, SGLang, KTransformers, or TensorRT-LLM, or accessed through Moonshot's OpenAI/Anthropic-compatible API and several third-party inference providers.

## Benchmarks

- SWE-bench Verified (agentic, single attempt): 69.2% ([source](https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905))
- SWE-bench Multilingual (agentic, single attempt): 55.9% ([source](https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905))
- Multi-SWE-Bench: 33.5% ([source](https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905))
- Terminal-Bench: 44.5% ([source](https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905))
- SWE-Dev: 66.6% ([source](https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905))
- LiveCodeBench v6 (Pass@1): 53.7% ([source](https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905))

## Pricing

- Input: $0.60 / 1M tokens per 1M tokens
- Output: $2.50 / 1M tokens per 1M tokens

Representative listed price; open weights can also be self-hosted, and rates vary across inference providers.

Source: https://openrouter.ai/moonshotai/kimi-k2-0905

## Strengths

- Agentic coding: improved accuracy and generalization across coding scaffolds, with strong SWE-bench Verified and SWE-Dev scores
- Frontend and UI generation: more aesthetic and functional output for web, 3D, and interface tasks
- 256K-token context for multi-file edits, large-repo Q&A, and long-horizon agent tasks
- Open weights under a permissive Modified MIT License, allowing self-hosting and commercial use
- Efficient MoE design: trillion-parameter capacity with only 32B parameters active per token
- Strong tool-use and function-calling behavior for building autonomous agents

## Best for

- Autonomous coding agents that edit, build, and test across multi-file repositories
- Frontend and web UI generation, including 3D and interactive layouts
- Repository-scale question answering and code understanding over large codebases
- Tool-calling and function-calling workflows for agent frameworks
- Self-hosted open-weight deployments where data control or cost predictability matters
- Long-context document and codebase processing up to 256K tokens

## How to access

- [Moonshot AI (Kimi) API](https://platform.moonshot.ai) — model id `kimi-k2-0905-preview`
- [OpenRouter](https://openrouter.ai/moonshotai/kimi-k2-0905) — model id `moonshotai/kimi-k2-0905`
- [Groq](https://console.groq.com/docs/model/moonshotai/kimi-k2-instruct-0905) — model id `moonshotai/kimi-k2-instruct-0905`
- [DeepInfra](https://deepinfra.com/moonshotai/Kimi-K2-Instruct-0905) — model id `moonshotai/Kimi-K2-Instruct-0905`

## Kimi K2 versions

- [Kimi K2.7-Code](https://ai-tldr.dev/models/kimi-k2-7-code/) (12 June 2026)
- [Kimi K2.6](https://ai-tldr.dev/models/kimi-k2-6/) (20 April 2026)
- [Kimi K2.5](https://ai-tldr.dev/models/kimi-k2-5/) (27 January 2026)
- Kimi K2-Instruct-0905 (9 September 2025) — this page
- [Kimi K2](https://ai-tldr.dev/models/kimi-k2/) (11 July 2025)

## FAQ

### What is Kimi K2-Instruct-0905?

It is an updated checkpoint of Moonshot AI's open-weight Kimi K2 model, released on September 9, 2025. It is a 1-trillion-parameter Mixture-of-Experts model with 32 billion active parameters per token, tuned for agentic coding and tool use, with weights published under a Modified MIT License.

### What changed in the 0905 update?

The 0905 release improves agentic coding accuracy and generalization across scaffolds, enhances frontend and UI code generation (web, 3D, and interface output), and extends the context window from 128K to 256K tokens for multi-file edits and long-horizon agent tasks.

### Is Kimi K2-Instruct-0905 open source?

The model weights are openly released on Hugging Face under a Modified MIT License, which permits commercial use and self-hosting. It can be run on inference engines such as vLLM, SGLang, KTransformers, and TensorRT-LLM.

### Does Kimi K2-Instruct-0905 support images or vision?

No. It is a text-only model. Inputs and outputs are text, and it is positioned for coding, tool use, and agentic workflows rather than multimodal tasks.

## Links

- [Hugging Face model card](https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905)
- [Kimi K2 GitHub repository](https://github.com/MoonshotAI/Kimi-K2)
- [Kimi K2 project page](https://moonshotai.github.io/Kimi-K2/)
- [Moonshot AI API platform](https://platform.moonshot.ai)
- [OpenRouter pricing & benchmarks](https://openrouter.ai/moonshotai/kimi-k2-0905)
- [Groq model docs](https://console.groq.com/docs/model/moonshotai/kimi-k2-instruct-0905)
- [Artificial Analysis benchmark profile](https://artificialanalysis.ai/models/kimi-k2-0905)

Source: AI/TLDR LLM registry, https://ai-tldr.dev/models/kimi-k2-instruct-0905/