moonshotai logo

Moonshot AI

Kimi K3

Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI.

At a glance

Input / 1M
$0.90
Output / 1M
$14
Context
1.05M
Max output
944K
Intelligence
43.6
Released
Jul 2026
Text inputImage inputVideo inputReasoningTool calling

Pricing

Kimi K3 API pricing

Per-token rates, plus what common tasks actually cost.

Input tokensPer 1M tokens$0.90
Output tokensPer 1M tokens$14
Cached input (read)Per 1M tokens$0.60
Cache writePer 1M tokensNot available

Chat message

2K in, 500 out

$8.80

per 1,000 requests

Coding task

30K in, 4K out

$83

per 1,000 requests

Long document summary

150K in, 2K out

$163

per 1,000 requests

Benchmarks

Kimi K3 benchmark scores

Independent scores from Artificial Analysis. Higher is better.

Intelligence index

43.6

Coding index

76.2

Agentic index

50

Specs

Context window and capabilities

What it can read, what it can write, and which API features it supports.

Context window
1.05M tokens
Max output
944K tokens
Input types
Image, Text, Video
Output types
Text
Reasoning effort
Max, High, Low
Release date
July 16, 2026
  • Tool calling
  • Structured outputs
  • JSON mode
  • Reasoning
  • Temperature
  • Stop sequences
  • Deterministic seed
  • Verbosity control

FAQ

Frequently asked questions

How much does Kimi K3 cost?

Kimi K3 costs $0.90 per 1M input tokens and $14 per 1M output tokens through the API. A typical chat message costs about $8.80 per 1,000 messages.

What is the context window of Kimi K3?

Kimi K3 accepts up to 1.05M tokens of input and can write up to 944K tokens in a single response.

What is the best alternative to Kimi K3?

GPT-6 Sol from OpenAI is a strong alternative. It offers: 1.4x cheaper output, higher intelligence score, larger 1.05m context.

Can I use Kimi K3 alongside other models?

Yes. Shortcut Chat sends one prompt to Kimi K3 and any other models you pick at the same time, so you can compare answers side by side.

moonshotai logo

Try Kimi K3 next to every other model.

Use Kimi K3 alongside all the other models. Use one at a time, or multiple together. Get the best answer from every AI.