Roar AI
DocsRequest access
All models
Moonshot AI · Open weights

Kimi K3

Moonshot AI’s 2.8-trillion-parameter open model, first at web research when it launched.

Input$3.705per 1M tokens
Output$18.525per 1M tokens
BillingOne invoice, in LKRwith every other model
Request early accessSee every model’s rate

About Kimi K3

Kimi K3 was billed as the first open model in the three-trillion class: 2.8 trillion parameters, with 104 billion working on each token. Its standout result is web research — 91.2% on BrowseComp, first at launch — and it placed third on the Artificial Analysis Intelligence Index, behind only Claude Fable 5 and GPT-5.6 Sol.

It reads images and video as well as text, holds a million tokens, and always thinks before answering. Running it yourself takes a cluster of 64 or more accelerators, which is why most people use it through an API.

Where it’s strong

Research agents that browse and read many sources.

Long coding agents.

Knowledge work across long documents and video.

Where to be careful

Always reasons, so there is no fast mode for simple requests.

Its launch scores mix Moonshot’s own tests with differing agent set-ups.

Moonshot’s licence asks very large products built on it to display “Kimi K3”.

Price in rupees

Input$3.705 per 1M tokens
Output$18.525 per 1M tokens
Cached input$0.351 per 1M tokens

For scale: 1,000 requests of about 3,000 tokens in and 1,000 out cost $29.64 — a long prompt and a page of answer each. Every rate is published and stays the same from one request to the next. See every model.

Specs

Made byMoonshot AI
ReleasedJuly 2026
TypeChat
Context window1.05M tokens
ReadsText, Images, Video
Thinks before answeringAlways on
WeightsOpen (Kimi K3 License)
Size2.8T total, 104B active
Model idkimi-k3

Using it from Sri Lanka

Kimi K3 is billed in rupees, on the same monthly invoice as every other model your team uses — no foreign card, no separate Moonshot AI account, and one key for all of it.

Prompts sent to Kimi K3 are processed outside Sri Lanka. If a workload has to stay in the country, Qwen3.8-27B (Colombo) runs on our own servers in Colombo.

Call it

Use the OpenAI SDK you already have — point it at Roar AI and name the model. Read the docs.

from openai import OpenAI

client = OpenAI(base_url="https://api.roar-ai.com/v1", api_key="roar_live_…")

reply = client.chat.completions.create(
    model="kimi-k3",
    messages=[{"role": "user", "content": "Summarise this contract in plain English."}],
)

Questions

How much does Kimi K3 cost in Sri Lanka?

Input is $3.705 per 1M tokens; output is $18.525 per 1M tokens. For scale, 1,000 requests of about 3,000 tokens in and 1,000 out cost $29.64. The rate is published and does not change from one request to the next.

Can I pay for Kimi K3 in Sri Lankan rupees?

Yes. Usage is billed to one monthly invoice in rupees, together with every other model your team uses, and you can also pay by card. There is no separate Moonshot AI account to open and no foreign card needed.

Does Kimi K3 keep data in Sri Lanka?

No — prompts sent to Kimi K3 are processed outside Sri Lanka. If your data has to stay in the country, use Qwen3.8-27B (Colombo), which runs on our servers in Colombo.

Is Kimi K3 open source?

Its weights are open under the Kimi K3 License licence, so you could run it yourself. Through Roar AI you call it like any other model, with no servers to manage.