Kimi AI

Moonshot AI's assistant: Kimi K3 with a 1M-token context, and an Agent that delivers slides, documents, spreadsheets and websites. A free mini-course from basics to power moves.

What it nails

Watch out for

Kimi is the AI assistant from Moonshot AI, on kimi.com, mobile and the Kimi Work desktop app. Its flagship Kimi K3 (July 2026) has a 1,048,576-token context and understands text, images and video natively. Kimi's signature is the Agent, which "autonomously builds an internal knowledge system for how to complete a complex task" and hands back "editable Office files, deployed websites, or reports".

Start here: the basics

  1. Open kimi.com or the Kimi app and sign in.
  2. Pick a model: K2.6 for "Fast dialogue and Q&A", K3 for "the strongest overall capability", K3 Cluster for large-scale search.
  3. Pick the reasoning effort: Standard, Advanced or Extreme.
  4. Write clear instructions. Kimi's guide: "The model can't read your mind."
How to add numbers in Excel?

Example by Moonshot AI · source ↗ · Kimi's official weak example.

How do I sum a row of numbers in an Excel table? I want to automatically sum each row in the entire table and place all the totals in the rightmost column named 'Total.'

Example by Moonshot AI · source ↗ · The better version from the official guide.

More rules from the official guide

A bat and a ball cost $1.10 together. The bat costs $1.00 more than the ball. How much is the ball?

Example by DataCamp · source ↗ · Made with Kimi K3. It reasons its way to the right answer: $0.05.

Look at this image. First describe what's in it in one paragraph. Then write a 4-line haiku inspired by it. Return your answer as JSON with keys 'description' and 'haiku'.

Example by Analytics Vidhya · source ↗ · This uses an image: attach one first. Made with Kimi K3.

Plans

Kimi can be used free. Paid memberships add Agent uses from one shared credit pool (official prices in CNY):

PlanPriceAgent uses per month
Andante¥49 / monthAbout 30
Moderato¥99 / monthAbout 60, plus Agent Swarm
Allegretto¥199 / monthAbout 150, plus Goal Mode
Allegro¥699 / monthAbout 360, plus million-token conversations

"Generating a simple PPT may use about 1–2% of your credits, while running one Deep Research task may use about 5–10%." Failed tasks cost nothing.

Power moves

1. Agent: finished work, not just answers

The Agent has 20+ built-in tools for websites, slides, documents, spreadsheets and deep research. Kimi's tips: give "clear background information and constraints", do not press stop even if the page looks frozen, and split big jobs "into 2–3 stages".

Create a professional PowerPoint presentation on the H-1B visa policy. The presentation should include:

- Introduction to H-1B Visa
- What the H-1B visa is, its purpose, and its eligibility criteria.
- Historical Trends (Last 10 Years)
- Annual allotment of H-1B visas over the past decade.
- Demographics of recipients (nationalities, age groups, gender).
- Skills and industries most commonly associated with H-1B approvals.

Use clear graphs and charts (line graphs, bar charts, pie charts) to showcase these trends.

- Latest H-1B Policy Updates: Key changes in the new policy, Government or regulatory rationale behind these updates, Impact Analysis, Potential effects on different demographics (by nationality, gender, experience level), Skills and industries most likely to be affected, Projected numbers and distribution of impacted groups.
- Summary of key insights.
- Strategic implications for professionals, employers, and policymakers.

Ensure the PPT is visually engaging, data-driven, and easy to follow, with consistent design, infographics, and citations where needed.

Example by Analytics Vidhya · source ↗ · Made with Kimi's agent (then called OK Computer). The picture is the first slide.

2. Agent Swarm: up to 300 helpers at once

Agent Swarm can "automatically generate and coordinate up to 300 sub-agents to work in parallel" across up to 4,000 steps. Ask for parallel work explicitly.

Build a fully playable Bubble Shooter game that runs locally in a web browser. The game should support mouse and touch aiming, show a live aim guideline, fire colored bubbles from the bottom center, bounce shots off side walls, and snap bubbles to a grid at the top on collision. Implement core gameplay logic including popping groups of three or more connected bubbles of the same color, dropping any disconnected clusters, tracking and displaying score, showing a next-bubble preview, and ending the game when bubbles cross a danger line near the bottom. Keep visuals simple but responsive, include a restart button, and finish with brief instructions on how to run and test the game locally.

Example by DataCamp · source ↗ · Made with Kimi K2.5 Agent Swarm.

I want you to research best practices for deploying open-source LLMs in production. Focus on real deployment patterns rather than theory. If possible, have different agents look into, Inference stacks like vLLM, TGI, llama.cpp, Quantization strategies and hardware trade-offs, and cost control techniques. I want a short, structured guide with common architectures, trade-offs, and concrete recommendations for small teams vs large-scale deployments.

Example by DataCamp · source ↗ · Made with Kimi K2.5 Agent Swarm.

3. Open weights and Kimi Code

K3's weights are published under the Kimi K3 License (model-as-a-service businesses earning over $20M in 12 months need a separate agreement). Kimi Code brings the model to your terminal.

Generate an SVG of a pelican riding a bicycle

Example by Simon Willison's Weblog · source ↗ · A famous informal test. Made with Kimi K3.

Moonshot lists K3's own limits: unstable output after switching models, and "excessive proactiveness", where it may decide more than you asked. Agent tasks usually take 5 to 20 minutes and can forget early details in long conversations. Review the privacy policy before sharing business data.

FAQ

Is Kimi free?

Yes, Kimi can be used free on the web and apps. Paid memberships add more Agent uses and features.

What is Kimi K3?

Moonshot AI's flagship model from July 2026: 2.8 trillion parameters, a 1M-token context, and native image and video understanding.

What can the Kimi Agent make?

Slides, documents, spreadsheets, PDFs, deployed websites and deep research reports.

What is Agent Swarm?

A mode where Kimi splits a big task across up to 300 sub-agents working in parallel.

Is there a thinking mode?

The separate thinking mode was merged into the model and reasoning-effort choices: Standard, Advanced or Extreme.

Sources: Kimi K3 blog · Model and mode selection · Membership pricing · Agent overview · Prompt best practices