Kimi K3.1 Pro

Long-horizon coding, 3D creation, and financial research with 1M context

Try now

Read Docs

About model

Kimi K3.1 Pro has 3T parameters and a 1M-token context window. On Together AI it costs $2.00 per 1M input tokens and $10.00 per 1M output tokens.

Kimi K3.1 Pro is the first model in Moonshot AI's K3.1 series, which Moonshot describes as its most capable 3T-class open-weight model to date. It is a post-trained version of Kimi K3, trained with substantially larger post-training compute through joint reinforcement learning across hundreds of downstream domains, with Moonshot reporting consistent gains in every domain. The model covers the full range of coding work, from quick fixes in real codebases to full-project builds and long-horizon engineering, and brings stronger composition and polish to front-end pages, Three.js scenes, and games built with Unity, Unreal Engine, and Godot. It infers shape, proportions, and spatial relationships from a text brief, sketch, or reference image to produce usable 3D models, and handles complete financial workflows with targeted improvements to factual accuracy. The model supports a 1M-token context window. Available on Together AI.

‍

Open-Weight Model

3T-class

Post-trained from Kimi K3 with substantially larger RL compute

Context Window

Full projects, long financial documents, and extended agent sessions

Parametric Modeling

Builds usable 3D models from text briefs, sketches, or reference images

Model key capabilities

Full-Range Coding: From quick fixes in real codebases to full-project builds, with polished front-end pages, Three.js scenes, and games across Unity, Unreal Engine, and Godot
3D Modeling & Design: Writes and runs scripts for Blender and CAD software to build parametric models, from mechanical parts and architecture to configurable products
Financial Research: End-to-end financial workflows over real documents, from modeling and investment analysis to auditable, source-grounded research
Production-Ready Infrastructure: 99.9% SLA, available on serverless and dedicated infrastructure

API usage

cURL
Python
Typescript

Endpoint:

moonshotai/Kimi-K3.1-Pro

curl -X POST "https://api.together.ai/v1/chat/completions" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/Kimi-K3.1-Pro",
    "messages": [
      {
        "role": "user",
        "content": "What are some fun things to do in New York?"
      }
    ]
}'

from together import Together

client = Together()

response = client.chat.completions.create(
  model="moonshotai/Kimi-K3.1-Pro",
  messages=[
    {
      "role": "user",
      "content": "What are some fun things to do in New York?"
    }
  ]
)
print(response.choices[0].message.content)

import Together from 'together-ai';
const together = new Together();

const completion = await together.chat.completions.create({
  model: 'moonshotai/Kimi-K3.1-Pro',
  messages: [
    {
      role: 'user',
      content: 'What are some fun things to do in New York?'
     }
  ],
});

console.log(completion.choices[0].message.content);

Model card
Architecture Overview:
• 3T-class open-weight model, per Moonshot AI, built as a post-trained version of Kimi K3
• 1M-token context window
• Accepts text and image input, including sketches, reference images, and multi-view renderings, with text output

Training Methodology:
• Substantially larger post-training compute than Kimi K3, with large-scale joint reinforcement learning across hundreds of downstream domains
• Moonshot optimized the full training stack, from the RL framework and long-context inference to sandbox infrastructure, rollout scheduling, and shared environment quality checks, so added RL compute delivers stable returns
• Targeted training to improve factual accuracy on real-world tasks and challenging prompts
• For the open-weight release, Moonshot identified and removed cyber-related data from both mid-training and RL stages while retaining general-purpose coding tasks

Performance Characteristics:
• Unless noted, results are Moonshot-reported at maximum reasoning effort
• Terminal coding (Terminal-Bench 4.0, realistic resource-constrained terminal tasks): 65.7%
• Long-horizon coding (Kimi Code Bench 3.0 Long-horizon, Moonshot in-house): 61.2%, up from 55.95% for Kimi K3
• 3D reconstruction (BenchCAD, CAD code generation from multi-view renderings): 92.7% with tools, 72.8% without
• Financial document analysis (OfficeQA Pro v2: 64.4, which Moonshot reports as state of the art; Frontier Finance investment workflows: 58.6)
• Moonshot in-house finance suite versus Kimi K3: financial modeling 57.6 from 50.4, financial research 70.4 from 58.8, business operation 37 from 8.9
• Factual accuracy (Kimi Hallu Bench Hard, Moonshot in-house stress test): claim-level hallucination rate 8.3%, down from 23.4% for Kimi K3
‍
Prompting
Together AI API Access:
• Access Kimi K3.1 Pro via Together AI APIs using the endpoint moonshotai/Kimi-K3.1-Pro
• Authenticate using your Together AI API key in request headers
• Published results use maximum reasoning effort with temperature 1.0 and top-p 0.95
• Supports text and image input across a 1M-token context window
• Available on Together AI serverless and dedicated infrastructure
‍
Applications & use cases
Agentic Software Engineering:
• Run long-horizon coding agents from quick fixes in real codebases to full-project builds
• Build front-end pages, Three.js scenes, and games across Unity, Unreal Engine, and Godot
• Diagnose and iterate on complex systems across many agent steps

3D Creation & Design:
• Generate parametric models for mechanical parts, architecture, and custom furniture
• Script Blender and CAD tools directly to build assets and adjust geometry, materials, and layout
• Create 3D assets for games, brand films, and interactive web experiences

Finance & Investment Research:
• Build financial models with linked formulas and sound financial logic
• Produce auditable investment research grounded in source documents
• Analyze complex fiscal data across large sets of real financial documents
‍