Kimi K3
Neural Network
Kimi K3 is a Moonshot AI reasoning model with a million-token context window for code, long documents and image analysis.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 1 048 576 tokens
- Max answer length
- 943 718 tokens
- Reasoning effort
- low, high, max
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Kimi K3?
Kimi K3 is a Moonshot AI model that reasons step by step before answering: a context window of 1,048,576 tokens and up to 943,718 output tokens let it hold an entire codebase or a whole document archive in view at once. Besides text it accepts images and documents, and it supports function calling with structured JSON output, so it fits agent pipelines and services that expect strict response schemas. In BotHub the model opens from Russia without a VPN and without a foreign card, and you pay in roubles for the tokens you actually spend. More than 250 models live in one window, so you can compare answers on the same prompt, while a single OpenAI-compatible API lets you swap engines without rewriting the integration; traffic is encrypted with AES-GCM and requests are not stored. Developers review large repositories and refactor legacy code with it, analysts pull figures out of long reports and scans, engineers build tool-calling agents on top of structured output, and research teams work through multi-step STEM problems.Frequently asked questions about Kimi K3
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
kimi-k3 from moonshotai handles text, reads documents and images, reasons before replying and calls external tools. With a context window of 1,048,576 tokens it fits long research, contract analysis, codebase review, refactoring, debugging and multi-step agent tasks in one session.
The limit is 943,718 output tokens per response, so a single answer can cover a full technical manual, a large refactoring diff or a detailed report. Real length depends on your prompt: ask for the volume and structure you actually need.
Yes. Reasoning is a confirmed capability: the model works through a task step by step before it answers, which helps on math, logic, STEM questions and debugging. Ask it to show the reasoning, and you can check each step yourself.
Yes, both. kimi-k3 supports function calling and structured JSON output, so you can wire it into agents, pipelines and services that expect a strict schema. Through BotHub's single OpenAI-compatible API you can switch to another model without rewriting the integration.
Images and documents are supported on input: send a screenshot, a diagram, a scan or a long PDF and ask questions about it. Audio and video input are not marked in our data. Output comes back as text.
Our data doesn't describe how the model behaves in specific languages, so we won't promise anything here; the honest answer is to test kimi-k3 on your own texts. BotHub itself works from Russia without a VPN, with payment by Russian cards.