Requesty
Moonshot AI logo
Moonshot AI

kimi-k3

Kimi K3 is Kimi’s most capable model to date, with 2.8 trillion parameters. Built on Kimi Delta Attention, a hybrid linear attention mechanism, and Attention Residuals, it offers native visual understanding and a 1M-token context window for frontier intelligence scenarios such as software engineering, knowledge work, and deep reasoning.

VisionReasoningTool callingCachingJSON schema
CompareDocs

Specifications

Context window1.0M tokens
Max output1.0M tokens
API typechat
AddedJul 16, 2026
Model IDmoonshot/kimi-k3
Data retentionYes
Used for trainingUnknown
Provider location🇨🇳 China
Privacy policyMoonshot Privacy Policy→

Benchmarks

Released 2026-07-16
Coding Indexcoding
76.2%

Artificial Analysis Coding Index — a composite of coding evaluations including LiveCodeBench, SciCode and Terminal-Bench.

GPQA Diamondreasoning
93.5%

Graduate-level physics, chemistry & biology questions designed to resist Googling.

Intelligence Indexreasoning
57.1%

Artificial Analysis Intelligence Index — a composite of multiple evaluations measuring overall model capability.

Scores are sourced from official model cards, Artificial Analysis, and public leaderboards. Benchmarks measure specific skills and do not capture every aspect of model quality. Always test on your own workload.

Pricing

Input / 1M
$3.00
Output / 1M
$15.00
Cache write / 1M
$15.00
Cache read / 1M
$0.30
Estimated cost
100K input + 10K output$0.45
1M input + 100K output$4.50
10M input + 1M output$45.00

Requesty charges exactly what the upstream provider charges, no markup, no per-request fees. Prompt caching and smart routing can reduce effective cost by 30-80%.

Quickstart

Drop-in compatible with the OpenAI SDK. Change the base URL, swap in your Requesty API key, and set the model to moonshot/kimi-k3.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
from openai import OpenAI

client = OpenAI(
    api_key="YOUR_REQUESTY_API_KEY",
    base_url="https://router.requesty.ai/v1",
)

response = client.chat.completions.create(
    model="moonshot/kimi-k3",
    messages=[
        {"role": "user", "content": "Explain quantum computing in one paragraph."},
    ],
)

print(response.choices[0].message.content)

Other Moonshot AI models

kimi-k2.7-code
262K ctx$0.95/1M
Coding 61
kimi-k2.6
262K ctx$0.95/1M
Coding 62
kimi-k2.5
262K ctx$0.60/1M
Coding 47

Frequently asked questions

How much does kimi-k3 cost?
kimi-k3 is priced at $3.00 per million input tokens and $15.00 per million output tokens when accessed via Requesty. Prompt caching is supported, which can cut effective input cost by up to 90% on repeated context. Requesty charges exactly what the upstream provider charges, we don't add markup.
What is the context window of kimi-k3?
kimi-k3 has a context window of 1.0M tokens, with a maximum output of 1.0M tokens per response. That's roughly 1,398 words of input you can fit in a single prompt.
How does kimi-k3 perform on benchmarks?
kimi-k3 scores 93.5% on GPQA Diamond, 76.2% on Coding Index, 58.7% on SciCode. See the full benchmark chart above for results across MMLU Pro, GPQA Diamond, SWE-Bench Verified, HumanEval, MATH, AIME, MMMU, and LiveBench.
What can kimi-k3 do?
kimi-k3 supports vision input, tool calling, extended reasoning, prompt caching, structured outputs (JSON schema). You can call it through any OpenAI-compatible client by pointing base_url to Requesty.
How do I use kimi-k3 with the OpenAI SDK?
Install the OpenAI SDK, set base_url to "https://router.requesty.ai/v1", set your API key to your Requesty key, and set the model to "moonshot/kimi-k3". The Quickstart above shows Python, JavaScript and cURL snippets.
Can I run kimi-k3 through Requesty?
Yes. kimi-k3 runs through Requesty's OpenAI-compatible API, served from Moonshot AI. You do not host the model yourself: point base_url at Requesty, set the model to "moonshot/kimi-k3", and requests are routed to the upstream provider with automatic failover. The same key gives you 600+ other models too.

Access kimi-k3 through Requesty

One API key, 600+ models, OpenAI-compatible. No markup on provider prices, automatic failover, and smart caching built-in.

All Moonshot AI models