Run Kimi K3.1 Pro in One Workspace
Send coding, 3D modeling and research briefs to K3.1 Pro and review every output in a single place
None

None

None

Long Story Video Skill

Turn a brief into a 10s–10min piece — the agent drafts the script, prompts and shots in one pass.

Ads Video Skill

Ads Video Skill

Build product ads end to end, with scripts, prompts and footage generated together.

3D Science Explainer Video Skill

Turn scientific concepts into 3D explainer animations

AI Video Prompt Generator

Feedback

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

  • ✓freeTrialImage.benefitHd
  • ✓freeTrialImage.benefitWatermark
  • ✓freeTrialImage.benefitUnlimited

kimi k3.1 pro

Try Kimi K3.1 Pro: an open-weight model that reads a million tokens at once and answers faster in its Highspeed mode for coding, CAD and research.

All Tools

Discover our comprehensive AI-powered animation toolkit

Kimi K3.1 Pro: Built for Heavy, Long-Running Workloads

A 3T-class open-weight release in the K3.1 line, offering a million-token working memory and a Highspeed option for quicker turnarounds.

  • Open Weights at 3T Scale, a Million Tokens Wide
    Debuting the K3.1 family, this is the strongest open-weight 3T-class model so far, holding a million-token working window for programming, 3D design and finance tasks.
  • Highspeed Mode That Cuts Wait Time
    Switch to the Highspeed build and replies land two to three times sooner than the standard model, making rapid back-and-forth iteration practical. Both builds ship today.
  • Reinforcement Learning Tuned Across Many Fields
    A post-trained K3 given substantially more training compute, refined from the RL stack through long-context inference, sandboxes and rollout scheduling for dependable gains.

Three Ways Into Kimi K3.1 Pro

Pick whichever entry point suits you — the web and mobile apps, the desktop tools, or the API and enterprise console.

What Kimi K3.1 Pro Handles Well

From shipping code and long autonomous runs to parametric CAD, financial analysis, factual accuracy and benchmark scores for K3.1 Pro.

Coding from Quick Patches to Whole Games

Expect everything from small fixes inside live repositories to complete project builds, with work across Unity, Unreal and Godot plus cleaner typography, animation and front-end detail.

Autonomous Runs with Self-Correction

In an internal walking-simulator trial covering 167 missions and 420 reviews, K3.1 Pro traced faults and fixed 36 shoulder moment-arm errors before retraining its own controller.

Parametric CAD and 3D Modeling

It reaches 92.7% on BenchCAD when tools are available and writes and executes Blender and CAD scripts on its own — no MCP connectors needed — for parts, buildings and bespoke furniture.

Financial Modeling and Traceable Research

OfficeQA Pro v2 comes in at 64.4, alongside gains over K3 of 7.2 points in financial modeling, 11.6 points in research and roughly 4.2x in simulated business operation.

Fewer Invented Claims on Real Work

Focused tuning brought claim-level hallucination on Kimi Hallu Bench Hard down from 23.4% to 8.3%, and the share of answers with at least one factual error fell to 42.4% of the earlier rate.

Benchmark Position and a Careful Open Release

Terminal-Bench 4.0 lands at 65.7%, above GPT-6 Astra's 58.2%, while removing cyber-related data moved xdev-bench-hard from 90.5% to 81.0%.

FAQ

Kimi K3.1 Pro: Your Questions Answered

What the context window covers, how the two builds differ, when the weights arrive and how it scores against closed models.

1

How much text can Kimi K3.1 Pro hold at once?

A million tokens fit in a single window. This is the opening model of the K3.1 line and the strongest open-weight 3T-class release so far.

2

What separates the standard build from Highspeed?

Highspeed responds two to three times quicker than the standard build, which suits fast iteration. Both are live now across Kimi.ai, Kimi Work, Kimi Code and the API.

3

When do the full weights land?

The complete weights for Kimi K3.1 Pro are scheduled for release by November 4, 2026, following published principles that keep offensive cyber capability below the frontier.

4

How does it stack up against closed frontier models?

On coding, agentic and knowledge tests it lands close to or level with the strongest proprietary systems: 65.7% on Terminal-Bench 4.0 against 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5, plus 92.7% on BenchCAD with tools.

5

Can finance teams rely on Kimi K3.1 Pro?

It posts 64.4 on OfficeQA Pro v2 and beats K3 in modeling and research, while claim-level hallucination dropped from 23.4% to 8.3% on Kimi Hallu Bench Hard — a deliberate stress test rather than an everyday error rate.

6

How do organizations switch it on?

Kimi Enterprise administrators can enable Kimi K3.1 Pro for their workspace; it ships switched off at launch. Kimi Business layers on enterprise data privacy, member management and separate personal and org accounts.

Put Kimi K3.1 Pro to Work Today

Kimi K3.1 Pro is available on Kimi.ai, the mobile apps, Kimi Work, Kimi Code and the API — enable it in Kimi Enterprise or pick Kimi Business for your team.