What is Moonshot AI?
Moonshot AI is a Chinese artificial intelligence company that builds the Kimi assistant and the Kimi K-series large language models. Its mission, in its own words, is to find the best conversion of energy into intelligence. Kimi handles chat, deep research, coding, slides, documents, spreadsheets and websites, and its agents can split large jobs into parallel tasks. Developers reach the same models through the Open Platform API or download open weights from Hugging Face.
The company was founded in March 2023 by Yang Zhilin, Zhou Xinyu and Wu Yuxin, Tsinghua University schoolmates, and is based in Beijing’s Haidian district, according to Wikipedia. Alibaba and Tencent are among its backers. Kimi launched in October 2023, Kimi K2 arrived as an open model in July 2025, and Kimi K3, with 2.8 trillion parameters, followed in July 2026. The service was live in October 2026.
Key Features
- Kimi K3 flagship model: Moonshot AI’s Kimi K3 is a 2.8 trillion parameter, natively multimodal model with a context window of 1,048,576 tokens on the API. The company aims it at coding, knowledge work and reasoning, and it powers the newest features in the Kimi app.
- Open-weight releases: Moonshot AI publishes Kimi K3 weights on Hugging Face, following the open release of Kimi K2 in July 2025. Developers can self-host, although Notebookcheck notes the full model needs data-centre hardware such as an Nvidia DGX B200.
- Agent and Agent Swarm: Kimi’s Agent mode completes multi-step tasks, and Agent Swarm splits large jobs across parallel sub-agents. One August 2026 review reports up to 300 sub-agents on the top tier, which suits broad research and batch drafting.
- Goal mode: Kimi’s Goal mode lets an agent keep working towards a stated objective with less step-by-step prompting. The official plan page lists it from the Allegretto tier at $39 a month upwards.
- Slides, docs and spreadsheets: Kimi produces presentations, documents and spreadsheet work as finished files, and the homepage promotes consulting-grade slides. Reviewers report exports to Word, PDF, Excel and PowerPoint, which saves a separate formatting step.
- Deep Research: Kimi’s Deep Research mode gathers sources and writes structured reports on a topic. TechRadar tested it alongside multi-document analysis and found the long-context handling a strength for research-heavy work.
- Websites and games: Kimi builds and deploys websites, and its homepage shows playable multiplayer and 3D games created from prompts. Visual coding tools help non-developers prototype interactive pages without setting up a local environment.
- Kimi Code: Moonshot AI’s Kimi Code is a coding tool included with every paid membership, according to the official overview. TechRadar notes VS Code and command-line agent integration, which places it alongside other terminal coding assistants.
- Kimi Claw and plugins: Kimi Claw brings the assistant into group chats, with 10 group chats on Allegretto and above. The app also offers plugins, scheduled tasks and memory features, and paid tiers support 15+ plugin types.
- Open Platform API: Moonshot AI sells API access through platform.kimi.ai, priced per million tokens. Kimi K3 costs $3.00 per million input tokens on a cache miss and $15.00 per million output tokens, with cheaper K2-series models for budget workloads.
Pros
- Leading open-weight model: Moonshot AI releases its flagship weights openly, which few frontier labs do. Notebookcheck reports Kimi K3 placed third on the Arena agent leaderboard, giving self-hosting teams a strong option they can inspect and run.
- Lower API prices: Kimi K3 costs less per token than many closed rivals on the official price list. K2-series models such as kimi-k2.6 start at $0.95 per million input tokens, which helps developers running high-volume agent workloads.
- Long context: Kimi K3 accepts over one million tokens of context through the API and Allegro chat. TechRadar praised Kimi’s long-context handling, which helps when analysing long contracts, codebases or several reports at once.
- Finished deliverables: Kimi turns prompts into slides, documents, spreadsheets and websites rather than plain chat replies. Reviewers found it useful for PDF-to-deck conversion and batch drafting, which shortens office tasks.
- Usable free tier: Kimi offers a free Adagio tier that third-party guides say needs no credit card. Chats with the K2.6 model reportedly do not consume credits, so casual users can test the assistant properly.
- Wide platform coverage: Kimi runs on the web, iOS, Android, macOS, Windows and a Chrome extension. You can start a task on a phone and continue it at a desk without switching tools.
Cons
- Data residency questions: Moonshot AI is a Beijing company, and TechRadar advises regulated teams to check its data policies first. Organisations with strict EU or US data sovereignty rules may need legal review before uploading sensitive files.
- Credit-based limits: Kimi meters agent work through monthly credits, from about 60 on Moderato to 720 on Vivace. Heavy Agent Swarm use can drain an allowance quickly, and one review describes high token use on the API.
- Best features cost more: Kimi locks Goal mode and Kimi Claw to Allegretto at $39 a month, and million-token chat to Allegro at $99. Free users get no Agent Swarm or K3 access, according to third-party pricing guides.
- Accuracy needs checking: Kimi can state wrong figures with the same confidence as correct ones, an August 2026 review warns. Statistics and citations in research outputs still need manual verification before you publish them.
- Subscription availability: Kimi’s paid tiers showed as sold out from 20 July 2026, according to one pricing guide. Availability may change, so confirm checkout works on the official page before planning around a plan.
- Distillation controversy: Moonshot AI faced allegations in February 2026 that it distilled Claude models through fraudulent accounts, according to Wikipedia. US committees later questioned firms using Kimi models, which may concern procurement teams.
Who is Using Moonshot AI?
- Developers: Call Kimi K3 through the API or self-host the open weights.
- Researchers and analysts: Run Deep Research and Agent Swarm across many sources.
- Office and consulting teams: Produce slides, documents and spreadsheets from prompts.
- Students and casual users: Chat with K2.6 on the free tier across web and mobile.
Pricing
Moonshot AI sells Kimi memberships from $19 to $199 a month, with annual billing cutting up to $480 a year, and bills its API separately per token (official prices, October 2026).
| Plan | Price | What you get |
|---|---|---|
| Adagio (free) | $0 | About 6 agent credits a month, one task at a time, unlimited K2.6 chat (as reported by third-party guides) |
| Moderato | $19/month or $180/year | 60 agent credits, 2 concurrent tasks, 4x priority, 2,000 professional database calls, Kimi Code |
| Allegretto | $39/month or $372/year | 150 agent credits, 5,000 database calls, Goal mode, Kimi Claw |
| Allegro | $99/month or $948/year | 360 agent credits, 4 concurrent tasks, 12,000 database calls, K3 chat up to 1M tokens |
| Vivace | $199/month or $1,908/year | 720 agent credits, 4 concurrent tasks, 24,000 database calls, Allegro features |
| API: kimi-k3 | $3.00 input (cache miss), $0.30 cache hit, $15.00 output per 1M tokens | 1,048,576-token context window |
| API: kimi-k2.6 | $0.95 input (cache miss), $0.16 cache hit, $4.00 output per 1M tokens | 262,144-token context window |
The Chinese site lists separate yuan tiers from ¥49 to ¥699 a month, one review quotes lower K2.6 API rates than the official list, and a pricing guide reported paid tiers as sold out from July 2026. Plans and quotas change often, so check the Kimi membership page before you subscribe.
What Makes Moonshot AI Unique?
Open weights at frontier scale set Moonshot AI apart. Kimi K3 is described as the largest open-weight model released so far, and the same model runs inside a consumer app that ships finished slides, documents, sites and games. Most rivals keep their top model closed or separate their chat and developer products more sharply. Moonshot AI pairs a downloadable model with Agent Swarm, Goal mode and low API prices, which appeals to teams that want both a ready product and control over hosting.
Compatibilities and Integrations
- Platforms: Web, iOS, Android, macOS, Windows and a Chrome extension.
- API: Open Platform API at platform.kimi.ai with K3 and K2-series models.
- Developer tools: Kimi Code with VS Code and command-line agent support.
- Files and exports: Word, PDF, Excel and PowerPoint input and output, plus website deployment.
Moonshot AI Tutorials
A first Kimi project takes a few minutes from sign-up to a finished file.
- Open kimi.com or the mobile app and sign in to start on the free Adagio tier.
- Pick a mode such as chat, Agent, Slides, Docs or Deep Research for your task.
- Describe the outcome you need and attach any PDFs, spreadsheets or images.
- Review the draft, ask for edits, then export or deploy the finished file or site.
- For developers, create a key on platform.kimi.ai and call kimi-k3 or kimi-k2.6 from your code.
How We Rated It
| Criterion | Score | Why |
|---|---|---|
| Accuracy and Reliability | 4.2 | Strong, but verify figures |
| Ease of Use | 4.1 | Clear modes, many options |
| Functionality and Features | 4.7 | Agents, slides, code, API |
| Performance and Speed | 4.1 | Fast, swarms use credits |
| Customization and Flexibility | 4.4 | Open weights, many modes |
| Data Privacy and Security | 3.2 | China-based data concerns |
| Support and Resources | 3.6 | Docs good, limited live help |
| Cost-Efficiency | 4.4 | Low API prices, free tier |
| Integration Capabilities | 4.3 | API, apps, Kimi Code |
| Overall Score | 4.1 | Average of the nine criteria |
Summary
Moonshot AI suits developers who want an open-weight frontier model and knowledge workers who want an agent that hands back finished files. The main limitation is trust: data residency questions and confident errors mean sensitive or published work needs care. Paid tiers add Goal mode, Kimi Claw and million-token chat, but credits run down quickly under heavy swarm use. Start on the free Adagio tier with K2.6, then test one real task in Agent mode before choosing a plan.
Reviews
Write a Review