Kimi K3.1 Pro Prime
High-speed Kimi K3.1 Pro from Moonshot for fast coding, reasoning, and image Q&A. Same model as K3.1 Pro, with a 1M-token context, tool calls, and web search.
## Overview
Kimi K3.1 Pro Prime is the high-speed version of Moonshot's Kimi K3.1 Pro, served on a faster lane. It uses the same model weights as K3.1 Pro, so quality is identical: a 1M-token context window, native image understanding, and strong results on code, financial research, and long-horizon agent tasks (65.7% on Terminal-Bench 4.0, 64.4 on OfficeQA Pro v2, and an 8.3% statement-level hallucination rate). It takes text and images and returns text.
## What you can build
- A fast coding assistant that explains errors from IDE screenshots and suggests fixes
- A screenshot-to-action agent that reads a UI and outputs step-by-step actions
- Image Q&A for product photos, charts, diagrams, and code inside screenshots
- Long-context work like reviewing large specs or logs pasted as text
- Tool-using assistants that call your functions for search, databases, or internal services
- Research answers grounded in fresh facts with web search
## Inputs
- A prompt in plain text, or a multi-turn conversation as messages
- Optional system instructions that set rules and style for the whole conversation
- Up to 10 images per request
- Optional user and session IDs to keep chat history across turns
- Optional reasoning effort: minimal (default), low, medium, high, xhigh, or max
- Optional output token limit, from 16 to 1,048,576, reasoning included
- Optional function tools, tool choice, and parallel tool calls for agent workflows
- Optional tool outputs from a previous turn to continue the same run
- Optional web search for questions that need current information
## Outputs
- The assistant's text answer, streamed as it is written
- Tool call requests with the tool name and JSON arguments, when tools are provided
- Reasoning kept separate from the final answer
## Recommended settings
- Keep minimal or low for everyday chat and quick answers.
- Use high or above for hard coding, math, and multi-step agent tasks.
- Turn on web search for recent events, prices, or anything that changes often.
## Limitations
- Image understanding depends on the input. Blurry screenshots, heavy compression, tiny fonts, and skewed scans increase errors.
- Only images are supported as attachments. Video, audio, and PDFs are not.
- Reasoning tokens are billed as output, so higher effort costs more.
- Each web search costs $0.01, and the search results are billed as input tokens.
- Answers without web search can be outdated.
## Safety & compliance
- Only upload images you have the rights to use.
- Don't submit personal data you don't have permission to process.
- Don't use the model for illegal activity, sexual content involving minors, non-consensual intimate imagery, or targeted hate and harassment.
- Follow applicable laws and platform usage rules.
API quick start
Run kimi k3.1 pro prime with a single API call.
{
"prompt": "Make a 7-day dinner plan for a family of …",
"inputImage": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}curl -X POST "https://api.wiro.ai/v1/Run/moonshot/kimi-k3-1-pro-prime" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "Make a 7-day dinner plan for a family of …",
"inputImage": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}
JSONOther models you might like
Explore similar llm & chat models.



