---
title: "Kimi K3 | Tarka Inference"
description: "Moonshot AI's flagship open model, for coding agents and long, multi-step tasks."
url: "https://tarkahq.com/inference/kimi-k3/"
---

Language model

# Kimi K3

Moonshot AI's flagship open model, for coding agents and long, multi-step tasks.

[Sign up](https://tarkahq.com/console?auth=signup) [Language model docs](https://tarkahq.com/docs/inference/llms/)

- **Status**: Available
- **Parameters**: 2.8T
- **Active**: 104B
- **Context**: 1M
- **Best for**: Coding agents, long multi-step tasks, and the hardest problems

## Model IDs

| Model ID | Use |
| --- | --- |
| `kimi-k3` | Coding agents and long, multi-step tasks |

Use one of the model IDs above with Chat Completions or Responses at `https://tarka.rest/v1`.

## Pricing

`kimi-k3`: $2.97 input · $0.297 cached input · $14.85 output per million tokens

 USD · catalog 2026-11-04.1. Release verified 2026-11-04. EUR uses the catalog conversion. [Live catalog](https://tarka.rest/control/v1/pricing).

## Overview

Kimi K3 is Moonshot AI’s open-weight model, released under the Kimi K3 License. It has 2.8 trillion parameters, 104 billion of them active for each token, and a context of about a million tokens. It always reasons before it answers.

## When to pick it

- Coding agents that plan, edit, and test across a whole repository
- Long tool loops where the model has to keep track of many steps
- Hard problems where the quality of the answer matters more than speed or price

## When to look elsewhere

- High-volume or latency-sensitive calls: start with GLM-5.3 Flash or DeepSeek V4.1 Flash
- Short, simple replies, where its reasoning adds time and tokens

## In a tool loop

Send each assistant message back exactly as you received it, including `reasoning_content` and `tool_calls`. Moonshot AI asks for the complete message to be passed back: the model needs its earlier reasoning to stay on track.