Kimi K3.1 Pro
moonshot/kimi-k3.1-pro
Back to model catalog

Moonshot AI · Model guide

Kimi K3.1 Pro

Compare
moonshot/kimi-k3.1-proReleased Oct 24, 2026

Moonshot's post-trained successor to Kimi K3: a natively multimodal 2.8T-parameter MoE with 104B active, a 1M-token context window and always-on reasoning, taking text, image and video input. Moonshot reports gains over K3 in coding, 3D creation, financial research and factual accuracy, and positions it for long-running agents.

Modalities
TextImageVideoText
Context
1M
Max output
1M
Price / 1M tokens
$ 2.00in$ 10.00out
StreamingTool callingJSON outputImage inputThinking mode

What this model is good at

What the vendor positions it for, and which of our models to reach for instead.

VendorMoonshot calls it its most capable 3T-class open-weight model to date, with substantial improvements in coding, 3D creation, financial research, factual accuracy and other agentic capabilities.source

Best for
  • Coding agents · The vendor reports Terminal-Bench 4.0 at 65.7 and DeepSWE v1.1 at 75.5, both run with the Claude Code harness.source

  • 3D modeling and CAD · Moonshot highlights 3D creation: from a brief, sketch or reference image it writes scripts for Blender and CAD tools, and it scores 92.7 on BenchCAD with tools.(vendor claim)source

  • Financial research · Moonshot targets end-to-end financial workflows and reports lower hallucination rates than K3 on its internal factual-accuracy stress test.(vendor claim)source

Neighbouring models
  • moonshot/kimi-k3predecessor —Choose K3 for workloads already validated on it: K3.1 Pro is post-trained from K3 with the same parameter count, context window and API parameters.source
Compare them side by side →

Pricing and billing

Billed per token. Cached input is charged at the cache-read rate.

Price / 1M tokensYou pay
Input$ 2.00
Output$ 10.00
Cache read$ 0.20
Cache write$ 2.00
Estimate a request
$ 0.0400
Estimated cost per request
10.0K × $ 2.00
+ 2.0K × $ 10.00
≈ $ 40.00 per 1,000 requests

Capabilities and limits

What CrossModel guarantees across every route this model can take right now.

Context window
1.0M tokens
Max output
1.0M tokens
Input / output modalities
Text + Image + Video → Text
Streaming
Supported
Tool calling
Supported
Structured output (JSON)
Supported
Image input
Supported
Thinking mode
Supported· always on
Reasoning effort
low · high · max
Available endpoints
/v1/chat/completions · /v1/responses · /v1/messages
Conditional limitsDisclosed by the vendor. These do not make the capability unavailable — they say when it is not.
  • tool_use.forced_choice is unavailable when thinking mode is on source

Published benchmarks

Scores the vendor reported, with the evaluation setup each one came from.

Agentic coding
  • Terminal-Bench 4.0
    pass_rate·Moonshot AI · 2026-10-24effort maxClaude Codesetup not fully disclosed

    厂商自测,使用 Claude Code 框架。同表中 Claude Opus 5.5 的 66.4 与 DeepSeek V4.1 Flash 的 42.3 与目录该身份的既有成绩一致。

    65.7%
Agentic search
  • BrowseComp (with tools)
    accuracy·Moonshot AI · 2026-10-24effort maxcompaction at 200K tokenssetup not fully disclosed

    上下文到 200K token 时压缩,最多 500 步,到上限时强制作答。采样参数与其它评测不同(top_p 1.0)。未披露具体工具集合。

    92.1%
Coding
  • DeepSWE 1.1
    resolved_rate·Moonshot AI · 2026-10-24effort maxClaude Codesetup not fully disclosed

    厂商自测,使用 Claude Code 框架;目录里 DeepSeek V4.1 Flash 同一身份的 77.0 用的是 Mini-SWE-V2,框架不同,直接比较时留意。

    75.5%
Computer use
  • OSWorld 2.1 (partial credit)
    score·Moonshot AI · 2026-10-24effort maxKimi Agentsetup not fully disclosed

    模型卡写作 OSWorld V2.1,使用官方 OSWorld v2 仓库中的 Kimi Agent 框架,未说明计分方式;同表中 Claude Opus 5.5 的 81.8 与目录该身份的既有成绩一致,据此归入部分计分这一身份。

    79.6%
Document understanding
  • OfficeQA Pro
    score·Moonshot AI · 2026-10-24effort maxClaude Codesetup not fully disclosed

    厂商自测,使用 Claude Code 框架;完整 PDF 语料只以页面图片提供,不含可机读文本。在线检索到参考答案的外部模型轨迹记零分。只给出百分数,保守记为 score。

    67.4%
Multimodal reasoning
  • MMMU Pro (no tools)
    accuracy·Moonshot AI · 2026-10-24effort max4 runssetup not fully disclosed

    按官方协议评测,保留原始输入顺序,图片置于文本之前;四次运行取平均。

    84.1%
Reasoning
  • Humanity's Last Exam Full (with tools)
    accuracy·Moonshot AI · 2026-10-24effort maxsetup not fully disclosed

    全集 2,500 题。未披露具体工具集合与框架。

    70%
Scientific reasoning
  • GPQA Diamond
    accuracy·Moonshot AI · 2026-10-24effort maxsetup not fully disclosed
    94.2%

Vendor-reported numbers, not CrossModel measurements. Scores are only comparable when the benchmark version, metric and evaluation setup match, so nothing here is averaged or ranked.

Use it in your tools

Point the base URL at CrossModel and paste this model ID — every tool below has a setup guide.

See all integration guides →

Frequently asked questions

What is Kimi K3.1 Pro?
Kimi K3.1 Pro is available on CrossModel as moonshot/kimi-k3.1-pro. It has a 1M-token context window and can return up to 1M tokens per request.
How much does Kimi K3.1 Pro cost?
Currently $ 2.00 per 1M input tokens and $ 10.00 per 1M output tokens. Cached input is billed at $ 0.20 per 1M.
Does Kimi K3.1 Pro support tool calling and structured output?
Tool calling: Supported · Structured output (JSON): Supported · Image input: Supported · Streaming: Supported · Thinking mode: Supported
Which endpoint do I call?
For text models, OpenAI Chat Completions (/v1/chat/completions), OpenAI Responses (/v1/responses) and Anthropic Messages (/v1/messages) all work — pick whichever your SDK already speaks.
Can I try it without writing code?
Yes. Sign in and open the Playground on this page to chat with the model directly — no API key needed. Usage is billed from your wallet at the normal rate.

Chat first, integrate later

Chat first, then wire it in once you like the answers.