What is Kimi — Moonshot AI's assistant explained

Kimi is an AI assistant made by the Chinese startup Moonshot AI — think ChatGPT or Claude, but built around very long context, autonomous "agent swarms" and unusually low pricing. You chat with it, upload and analyze long documents, search the web, write code and hand it multi-step tasks it runs on its own. You can use it free at kimi.com, and developers can call its models via one of the cheapest APIs on the market. Here's what Kimi actually is, what it does, and how it stacks up.

Key takeaways

  • What it is: Moonshot AI's chat assistant and model family (the "K-series"), a leading Chinese rival to ChatGPT and Claude.
  • Best known for: long context, its Agent Swarm (many agents running in parallel), the Kimi Code coding agent, and low cost.
  • Flagship model: Kimi K3 (July 2026) — open-weight, 1M-token context; earlier K2.5/K2.6/K2.7 Code stay available cheaper.
  • Free? Yes — the Adagio tier covers unlimited basic chat, uploads and web search; paid tiers add agent power. Here's how to start free.
  • Safe? Fine for everyday use, but it trains on your input and hosts data in Singapore — avoid sensitive data.

Who Makes Kimi?

Kimi is built by Moonshot AI, a Chinese AI startup founded in 2023 and now one of the country's most prominent large-model labs — often grouped with DeepSeek, Zhipu and MiniMax among China's leading AI companies. "Kimi" is the consumer-facing brand: the assistant you talk to, and the name of the underlying models (the K-series). If you've seen references to "Kimi K2" or "Kimi K3," those are Moonshot's models, the way GPT-5 is OpenAI's or Claude Opus is Anthropic's.

What Can Kimi Do?

At its core Kimi is a general assistant, and the basics are exactly what you'd expect:

  • Chat & reasoning — ask questions, brainstorm, write and edit text, get explanations.
  • Long documents — upload large files and have Kimi summarize, search and answer across them. Long context has been Kimi's signature strength since its early versions.
  • Web search — pull in current information rather than relying only on training data.
  • Coding — via Kimi Code, its coding agent, which paid plans scale up with a usage multiplier.

Where Kimi gets distinctive is its agent stack. Agent Swarm lets Kimi spin up many autonomous agents that work in parallel on a task — useful for large research or coding fan-out where you want dozens of things happening at once. Kimi Claw adds browser and cloud automation, so the assistant can act across the web and cloud tools rather than just chatting. These agent features are the reason Kimi's paid tiers exist — they're what you're really buying as you move up the ladder.

Kimi's Models: K3 and the K2 Series

Moonshot ships its capability under the K-series. The current flagship is Kimi K3, released in July 2026 — a large open-weight model with a 1-million-token context window. Below it sit the still-available K2.5, K2.6 and the coding-tuned K2.7 Code, which are cheaper and cover most everyday work.

Because the models are open-weight, they're not locked to Kimi's own app — you (or a cloud provider) can host them yourself, which is part of why Kimi's API is so cheap. On the API, K3 runs $3/$15 per 1M input/output tokens, while K2.5 is just $0.60/$3.00 — a fraction of what Western flagships cost.

Is Kimi Free?

Yes — Kimi has a genuinely usable free plan. Its five consumer tiers are named after musical tempos:

  • Adagio (Free) — unlimited basic chat, file uploads and web search, with limited autonomous-agent use.
  • Moderato ($19/mo) — entry agent tier: Kimi Code, Agent Swarm runs, more concurrency.
  • Allegretto ($39/mo) — bigger Kimi Code multiplier, Kimi Claw included.
  • Allegro ($99/mo) and Vivace ($199/mo) — heavy agent quotas and up to hundreds of parallel swarm agents.

For most people the free Adagio plan is plenty to try it as a chatbot and document tool. If you want the step-by-step, we wrote a guide to using Kimi for free, and a full plan-and-price breakdown against Claude.

Is Kimi Safe?

Two honest caveats, both about data rather than quality. First, Kimi's international platform states that it stores collected data on servers in Singapore and uses user input to train its models, with no clear opt-out — so don't paste confidential, client or regulated data into the hosted app. Second, as a Chinese model it applies content rules on China-related political topics — though, interestingly, independent testing in 2026 found Kimi censors far less than most Chinese models, closer to Western assistants on neutrality tests.

The practical takeaway is the same as for any Chinese model: great for everyday tasks, but for sensitive work use a Western assistant or self-host the open weights. We cover the full picture in are Chinese AI models safe?

Should You Use Kimi?

Use Kimi if you want a capable free assistant for chat and long-document work, you're into autonomous agents and want to experiment with agent swarms cheaply, or you're a developer who wants near-frontier models at rock-bottom API prices. Look elsewhere if you need the most mature ecosystem, the strongest privacy guarantees, or the absolute top of reasoning capability — that's where Claude and ChatGPT still lead.

Want to see how it compares directly? Read our Kimi vs Claude breakdown, or see where Chinese models sit in the broader 2026 AI API price war. And if you just want a recommendation for your specific use case, our free picker takes about 30 seconds.

Frequently Asked Questions

What is Kimi?

Kimi is an AI assistant made by the Chinese startup Moonshot AI. It works like ChatGPT or Claude — chat, long-document analysis, web search, coding and autonomous agents — but is known for long context, its Agent Swarm feature and low pricing. It's free to use at kimi.com.

Who owns Kimi?

Moonshot AI (Yuewen Group), a Chinese startup founded in 2023 and one of the country's leading large-model labs. Kimi is its consumer assistant and the brand for its K-series models.

Is Kimi free?

Yes. The free Adagio plan covers unlimited basic chat, file uploads and web search, with limited agent use. Paid tiers ($19–$199/mo) add agent credits, Kimi Code and Agent Swarm. The API is separately very cheap (from $0.60 per 1M input tokens).

What is Kimi K3?

Moonshot's flagship model, released July 2026 — a large open-weight model with a 1M-token context window, priced at $3/$15 per 1M tokens on the API. It succeeds the K2 series (K2.5, K2.6, K2.7 Code), which remain available cheaper.

Is Kimi safe to use?

Fine for everyday, non-sensitive tasks. Its international platform stores data in Singapore and trains on user input with no clear opt-out, so avoid confidential data. It censors politically sensitive topics less than most Chinese models. For sensitive work, prefer Claude or self-host an open-weight model.

Is Kimi better than ChatGPT or DeepSeek?

Depends on the job. Kimi wins on long context, agent swarms and cost; ChatGPT and Claude have more mature ecosystems (and Claude, stronger privacy); DeepSeek is an even cheaper Chinese rival on some tiers. For autonomous-agent workflows and cheap high-volume use, Kimi is compelling.