kimi k3.1 Build Console
Describe the job once — code, a 3D asset, a research deck or a whole site — and kimi k3.1 builds it; Highspeed cuts the wait while you iterate.
None

None

None

Long Story Video Skill

Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.

Ads Video Skill

Ads Video Skill

Generate professional ads and sales videos—AI auto-generates scripts, prompts, and footage.

3D Science Explainer Video Skill

Convert scientific concepts into stunning 3D explain animations

AI Video Prompt Generator

Feedback

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

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

kimi k3.1

Hand kimi k3.1 a brief and get code, 3D assets or a working page back — with a million-token memory and a faster Highspeed tier when time matters.

All Tools

Discover our comprehensive AI-powered animation toolkit

kimi k3.1: Deeper Memory, Quicker Turns

Kimi's first K3.1 release pushes the K3 base through far heavier post-training, and arrives with open weights at 3T scale plus a one-million-token window.

  • Trillion-scale weights, open to all
    The biggest open-weight drop in the K3.1 family so far, carrying roughly three trillion parameters and enough room for a million tokens of working memory.
  • A Highspeed tier that answers sooner
    Reach for the Highspeed build when latency matters — it returns results two to three times faster than the standard endpoint during back-and-forth work.
  • Joint RL over hundreds of fields
    Starting from K3 with much deeper post-training, the model runs large-scale joint reinforcement learning across hundreds of downstream domains, lifting each one.

Getting Started with kimi k3.1

Three short steps take you from a blank prompt to a published build.

What kimi k3.1 Can Actually Do

Coding agents, CAD and 3D creation, finance research and carefully bounded cyber limits — the ground kimi k3.1 covers.

Coding agents that stay on task

Terminal-Bench 4.0 lands at 65.7% — ahead of GPT-6 Astra (58.2%) and just shy of Claude Opus 5.5 (66.4%) — while Kimi Code Bench 3.0 Long-horizon climbs from K3's 55.95% to 61.2%.

Game engines and Three.js work

Unity, Unreal Engine and Godot are all supported alongside web search and 3D generation, and the front-end pages it writes show tighter layout, typography and motion.

CAD and parametric 3D scripting

BenchCAD with tools comes in at 92.7%; it also writes Blender or CAD scripts to reshape assets, swap parts and tune geometry, materials and layout — no MCP connectors required.

Finance work, start to finish

OfficeQA Pro v2 scores 64.4, and versus K3 it gains 7.2 points on Financial Modeling, 11.6 on Financial Research and about 4.2x on Business Operation.

Hallucination held down

On the internal Kimi Hallu Bench Hard, claim-level hallucination falls from 23.4% on K3 to 8.3%, and answers carrying at least one factual slip drop to 42.4% of K3's rate.

Cyber ability kept in bounds

Cyber data was stripped during mid-training and RL; the shipped model still clears 81.0% of xdev-bench-hard at pass@3, against 90.5% for the checkpoint without those filters.

FAQ

kimi k3.1: Questions Answered

Answers on what kimi k3.1 is, how the Highspeed build differs, when the weights land and where you can run it today.

1

So what is kimi k3.1?

It is the first model in Kimi's K3.1 series — the K3 base taken through much heavier post-training, released as open weights at 3T scale with room for a million tokens.

2

How far ahead of K3 is it?

Coding, agent and knowledge benchmarks all move up: Kimi Code Bench 3.0 Long-horizon rises from 55.95% to 61.2%, and hallucination drops from 23.4% to 8.3%.

3

When will the weights be downloadable?

Full weights are scheduled for release by Nov 4, 2026; hosted access through Kimi's own products is already available today.

4

Where can I use it right now?

Through kimi.ai and the mobile apps, Kimi Work 3.1.0 or newer, Kimi Code, and the Kimi API — which offers Zero Data Retention.

5

Is there a faster version?

Yes. Highspeed returns answers two to three times sooner and is callable as kimi-k3.1-pro-highspeed on the Kimi API Platform.

6

How does it stack up against closed frontier models?

It trades blows near the top: 65.7% on Terminal-Bench 4.0 versus Claude Opus 5.5 at 66.4% and GPT-6 Astra at 58.2%, plus 92.7% on BenchCAD with tools.

Start Building with kimi k3.1

Launch kimi k3.1 on Kimi.ai or through the API, drop into Highspeed whenever latency matters, and look for the full weights by Nov 4, 2026.