Moonshot AI · all models ›

MKimi K3OPEN

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

1MContext window · tokens
$3Input price · per M tokens
$15Output price · per M tokens
Moonshot AIProvider

Prices update automatically — checked daily against provider list prices.

See model comparisons → Compare all model prices

Benchmarks & performance

Independent benchmark scores for Kimi K3, measured by Artificial Analysis. Higher is better. Measurement mode: max.

Intelligence index59.7
Coding index76.2
GPQA93.5%
Humanity's Last Exam46.9%
Long Context Reasoning82.7%
SciCode58.7%
τ-Bench Banking46%
Terminal-Bench85%
⚡ Speed37.9 tokens/sec
⏱ Latency5.48s to first token
💰 Blended price$6 / 1M tokens
📈 Value10 intelligence points per $
vs. models measured heretop 3%
Scores higher than 97% of the 273 models measured by Artificial Analysis and tracked here.
Models at this level cost $3 per 1M tokens (median of 6) — this one costs $6.
Cheaper and better on this index: Claude Opus 5 (batch) · GPT-5.6 Sol · Grok 4.6 and 1 more
Benchmark data by Artificial Analysis

About this model

Kimi K3 is an open-weight AI model by Moonshot AI. You can download and self-host it for free; the prices below are hosted-API list prices, tracked daily, for when you prefer convenience over self-hosting.

Frequently asked questions

What is Kimi K3?

Kimi K3 is an AI language model from Moonshot AI. It is open-weight: you can download it and run it on your own hardware, for free. It scores 59.7 on the Artificial Analysis intelligence index.

Is Kimi K3 free?

The weights are free and open — you can self-host Kimi K3 and pay nothing per token. If you prefer a hosted API, list prices are $3 per million input tokens and $15 per million output tokens.

What is Kimi K3 good at?

Independent benchmarks from Artificial Analysis give it GPQA 93.5%, Humanity's Last Exam 46.9%, Long Context Reasoning 82.7%, SciCode 58.7%, τ-Bench Banking 46%, Terminal-Bench 85%. It is particularly used for code generation.

How fast is Kimi K3?

It generates about 37.9 tokens per second, with a median 5.48s delay before the first token. Measured independently by Artificial Analysis.

Can I self-host Kimi K3?

Yes. Kimi K3 has open weights, so you can download it and run it on your own GPU or server with tools like Ollama, vLLM or llama.cpp — with no per-token cost.

Related models

Kimi K2.6Moonshot AIKimi K2.5Moonshot AIKimi K2.7 CodeMoonshot AIKimi K2 0905Moonshot AIKimi K2 0711Moonshot AIKimi K2 ThinkingMoonshot AI