by Moonshot AI·Released 16 July 2026
#3 for websites · Kimi K3 reads text, pictures and video and writes text back, taking in about a thousand pages at once. You can download it and run it yourself, and it thinks before it answers. It is built for work inside large code projects: it uses connected tools, finds bugs, and revises its code from screenshots, log files, test results and what happens when the program runs.
You pay for what each request uses. No subscription, no minimum.
One message, the ordinary use
Every rate
| Meter | ↑ you send | ↓ comes back |
|---|---|---|
| Text tokens | 758,87 DA | 3 794,33 DA |
| Cached tokens read | 75,89 DA |
per 1M tokens
One company runs this model. OpenDunes sends every request to it. DA per 1M tokens: ↑ what you send · ↓ what comes back.
| Provider | ↑ Sent | ↓ Back | Cache read | Latency | Throughput | Uptime |
|---|---|---|---|---|---|---|
| DeepInfraUnited States | 758,87 DA | 3 794,33 DA | 75,89 DA | … | … | … |
You pay the price of the provider that served your request. Prices read 15:19 · speed over the last 30 minutes · uptime over 3 days.
Swap the base URL and the key. Your OpenAI code runs unchanged.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://opendunes.com/v1",
api_key=os.environ["OPENDUNES_API_KEY"],
)
resp = client.chat.completions.create(
model="moonshotai/kimi-k3:batch",
messages=[{"role": "user", "content": "Two lines to welcome a shop in Oran."}],
)
print(resp.choices[0].message.content)Runs this request once on your balance.
Standard OpenAI fields, grouped by what they control. Anything not listed is accepted and ignored.
Shaping the reply
How long, how varied, where to stop.
max_tokenstemperaturetop_ptop_kmin_pstopseedfrequency_penaltypresence_penaltyrepetition_penaltylogit_biasReasoning
Let it think before answering — and see the thinking.
reasoningreasoning_effortinclude_reasoningStructured output
Get JSON you can parse instead of prose.
response_formatstructured_outputsTools
Let the model call your functions.
toolstool_choiceAbout 2,66 DA per message. ↑ 758,87 DA · ↓ 3 794,33 DA per 1M tokens.
No. You pay per use from your balance, with no subscription. Run it once from Call it in code, at the same price.
#3 for websites · Kimi K3 reads text, pictures and video and writes text back, taking in about a thousand pages at once. You can download it and run it yourself, and it thinks before it answers. It is built for work inside large code projects: it uses connected tools, finds bugs, and revises its code from screenshots, log files, test results and what happens when the program runs.
Deposit dinars with your Edahabia or CIB card. Every model, Kimi K3 (batch) included, is paid from that balance. No foreign card, no conversion.
Moonshot AI names no languages for Kimi K3 (batch). Arabic, French and Darja: not stated.
It reads text, images and video. Up to 1M tokens in one request, about 1 000 pages. It writes up to 12 000 words at once.
Yes, its weights are public. Open weights on Hugging Face
Moonshot AI. Released 16 July 2026.
#1 for interface design · This model accepts text, pictures, and video, allowing it to read up to 1000 pages of information at once. It is designed to handle complex, long-running coding and reasoning tasks. A key feature is its ability to process visual and auditory data alongside written text. However, it is limited to generating text responses and cannot search the internet.
#3 for websites · Kimi K3 reads text, pictures and video and writes text back, taking in about a thousand pages at once. You can download it and run it yourself, and it thinks before it answers. It is built for work inside large code projects: it uses connected tools, finds bugs, and revises its code from screenshots, log files, test results and what happens when the program runs.
#2 for slide decks · Kimi K2.6 reads text and pictures and writes text back, keeping about 256 pages in mind, and it thinks before it answers unless you switch that off for quicker replies. It is aimed at long coding runs across Go, Rust and Python projects and at turning a written description or a mock-up into a working screen, and it can split a job across hundreds of helper copies of itself working side by side.
The coding-focused Kimi K2, which always thinks and keeps its full reasoning from turn to turn, reading text and images over 256K for end-to-end programming and breaking agent tasks into steps
From what its maker and two benchmark sites publish, and where it stands on OpenDunes. Read 7 Nov 2026.
Thinks first
Works through the problem before it answers.
Uses your tools
Can call functions your code provides.
Fills your JSON
Answers in the JSON shape you give it.
Searches the web No.
Can look things up online while it answers.
Cheaper on repeats
A prompt it saw a moment ago costs less the second time.
Open weights
The model's weights are public.
Debug a failing build from its logs and fix it
Thinks first
Navigate a large repository and make a cross-cutting change
Run a research agent that uses your tools and writes a report
Uses your tools