Moonshot AI

Available Day 0 Chat Reasoning Vision

Kimi K3.1 Pro

Moonshot AI's most capable open-weight model, for long-horizon coding, agents and research, with a 1M-token context window.

Start building Try in Playground Read the docs

Tokens of context: 1M
in, text out: Text + image
Input / output per 1M tokens: $2.00 / $10.00
Cached input per 1M tokens: $0.20

About the model

Kimi K3.1 Pro is the first model in Moonshot AI's K3.1 series, and its most capable 3T-class open-weight model to date. It's a post-trained version of Kimi K3, with substantially more reinforcement learning, and keeps K3's 1M-token context window and native image understanding, with configurable reasoning effort.

Start building

OpenAI compatible: point your code or coding agent at Verda Inference with your API key and the model ID kimi-k3-1-pro .

curl -X POST "https://inference.verda.com/v1/chat/completions" \
  -H "Authorization: Bearer ${VERDA_INFERENCE_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k3-1-pro",
    "messages": [
      {
        "role": "user",
        "content": "What is the best viewpoint in Helsinki?"
      }
    ],
    "max_tokens": 2048
  }'

Reasoning effort is configurable with reasoning_effort , from none to max. The default is max.

See the docs

What it's good at

Long-horizon coding

From quick fixes in real codebases to full-project builds, including games in Unity, Unreal Engine and Godot, sustained over many steps of tool use.

Use it for: coding agents that plan, build, test and revise.

Agentic work

Plans, calls tools and carries out multi-step tasks across many domains, from research to automation, with a much lower hallucination rate than Kimi K3.

Use it for: agents that plan, use tools and work through long tasks.

Financial research

Investment and fiscal analysis end to end, across real financial documents, with stronger factual accuracy on real-world tasks.

Use it for: research and analysis over filings, reports and data.

3D creation and design

Turns a brief, sketch or reference image into usable 3D models, writing and running Blender and CAD scripts to build parametric assets.

Use it for: design tools, game assets and product configurators.

Benchmarks

Quality checks run by Moonshot AI on our deployment, followed by benchmark results published by Moonshot AI, at maximum reasoning effort.

Verified on Verda by Moonshot AI

Kimi Tool Call Tool-call triggering and JSON schema accuracy

1.000
OCRBench Reading text in images

0.878
MMMU-Pro Vision Reasoning over images across subjects

0.846
BEAM (1M) Memory across conversations of about 1M tokens

0.3976
DeepSWE Coding-agent tasks in real repositories

0.757

As reported by Moonshot AI

Coding

Terminal-Bench 4.0 65.7
Kimi Code Bench 3.0 Long-horizon 61.2
SWE Marathon 51.7

Agentic

OSWorld V2.1 79.6
HLE Full with tools 70.0
Agents’ Last Exam 43.2

Finance and documents

OfficeQA Pro v2 64.4
Frontier Finance 58.6
GDP.pdf 34.6

Vision and 3D

BenchCAD with tools 92.7
Chartography with tools 90.2
Zerobench with tools (pass@5) 63.0

Try other models

DeepSeek

Coming soon

V4.1 Flash

A fast, low cost model for high-volume, everyday tasks.

Context: 1M tokens
Input: Text & image
Tool calling: Supported

z.AI

Coming soon

GLM 5.3

Strong coding and agent performance at a lower cost, for long-running tasks.

Context: 1M tokens
Input: Text
Tool calling: Supported

Available now

Start building with Verda Inference

Make your first call in minutes.

Start building or contact sales

One platform, the full AI lifecycle

From rapid prototyping to foundation training and scalable inference — on a single full-stack AI cloud

Start building Talk to an expert