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
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
Feedback
freeTrialImage.bannerPity
freeTrialImage.upgradeUnlock
- ✓freeTrialImage.benefitHd
- ✓freeTrialImage.benefitWatermark
- ✓freeTrialImage.benefitUnlimited
kimi k3.1 pro
Kimi K3.1 Pro brings a million-token window, open weights and a Highspeed mode that replies two to three times faster.
All Tools
Discover our comprehensive AI-powered animation toolkit
MiniMax H3
MiniMax H3 AI Video Generator
Seedance 2.5
The Future of AI Video Is Here.
Seedance 2.0
The Future of AI Video Is Here.
Veo3.1
Create Stunning Videos with Veo3.1
FLUX 3 Video Generator
Kling 3.0
Next-Gen AI Video Generator
Grok Video Generator
Create Videos from Text or Images with AI
MiniMax H3 video generator
MiniMax H3 AI Video Generator
Inside Kimi K3.1 Pro: Scale, Speed and Open Weights
A 3T-class open-weight release that leads the K3.1 line, pairing a one-million-token window with a faster Highspeed option.
- Million-Token Window on Open 3T-Class WeightsLeading the K3.1 line, this release carries a one-million-token window and enough capacity to hold coding, 3D and finance projects in a single pass.
- A Highspeed Mode That Answers 2-3x SoonerSwitch to the Highspeed build when you want tighter back-and-forth: replies land two to three times sooner than the standard model, and both builds ship today.
- Reinforcement Learning Tuned Across Hundreds of DomainsPost-training was rebuilt end to end — RL framework, long-context inference, sandboxes and rollout scheduling — on far more compute, so the gains hold across domains.
Ways to Access Kimi K3.1 Pro
Three entry points: the web and mobile apps, the desktop coding tools, or the API and an enterprise rollout.
Capabilities Inside Kimi K3.1 Pro
From full codebases and game engines to parametric CAD, finance research and hallucination control — here is where the model earns its place.
Whole-Project Coding, Game Engines Included
Handles small fixes in live repos and builds entire projects, with working knowledge of Unity, Unreal and Godot plus sharper typography, animation and front-end finish.
Multi-Step Autonomy With Self-Correction
Across an internal walking-simulator run of 167 missions and 420 reviews, the model traced faults and fixed 36 shoulder moment-arm errors before retraining its own controller.
Parametric CAD and 3D Modeling
Scores 92.7% on BenchCAD with tools and writes plus executes Blender and CAD scripts directly — no MCP connectors — for parts, buildings and bespoke furniture.
Finance Workflows and Traceable Research
64.4 on OfficeQA Pro v2, alongside gains over K3 of 7.2 points in financial modeling, 11.6 in research and roughly 4.2x in simulated business operation.
Fewer Invented Claims on Live Work
Focused tuning brought claim-level hallucination on Kimi Hallu Bench Hard from 23.4% down to 8.3%, and answers containing at least one factual slip now sit at 42.4% of the earlier rate.
Benchmark Results and Guarded Open Weights
Terminal-Bench 4.0 lands at 65.7%, ahead of GPT-6 Astra at 58.2%, while removing cyber-related data moved xdev-bench-hard from 90.5% to 81.0%.
Kimi K3.1 Pro: Questions Answered
Straight answers on window size, Highspeed mode, weight release dates, benchmark comparisons and enterprise access.
What context window does Kimi K3.1 Pro offer?
It carries a one-million-token window. This is the opening release in the K3.1 line and the strongest 3T-class open-weight model shipped so far.
How does Highspeed differ from the standard model?
Highspeed responds two to three times faster, which suits rapid back-and-forth. Both builds are live now on Kimi.ai, Kimi Work, Kimi Code and the Kimi API.
When do the full model weights land?
Full weights are scheduled for Nov 4, 2026, released under stated principles that keep offensive cyber capability below the frontier.
How does it stack up against closed frontier models?
On coding, agentic and knowledge tests it matches or nears the best closed 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.
Can finance teams rely on its accuracy?
OfficeQA Pro v2 gives 64.4, with modeling and research both improved over K3; claim-level hallucination on Kimi Hallu Bench Hard fell from 23.4% to 8.3% — that bench is a deliberate stress test, not an everyday error rate.
How can an organization switch it on?
Enterprise admins enable it for their workspace; access is off by default at launch. Kimi Business layers on data privacy, member management and split personal and org accounts.
Start Building With Kimi K3.1 Pro
Available now across Kimi.ai, the mobile apps, Kimi Work, Kimi Code and the API. Turn it on in Kimi Enterprise, or pick Kimi Business to roll it out team-wide.