Kimi AI is the family of AI products from Moonshot AI: a chat assistant at kimi.ai and in mobile apps, a desktop app called Kimi Work, a coding tool called Kimi Code, a pay as you go API, and, for its newest model, Kimi K3.1 Pro, open weights anyone can download. When people say "Kimi," they might mean any of those, so the useful question isn't only what Kimi is. It's which Kimi fits what you want to do.
This guide maps the product family, says what each piece is for, and flags the caveats that matter before you build on it: the license, the hardware needed to self host, and which numbers come from Moonshot itself.
Who makes Kimi
Kimi is built by Moonshot AI. Its current flagship model is Kimi K3.1 Pro, which Moonshot introduced in late October 2026 as the first model in its K3.1 series and, in its words, its most capable 3T class model to date. Moonshot also released K3.1 Pro Highspeed, which it says runs two to three times faster.
According to Moonshot's model card on Hugging Face, K3.1 Pro is a mixture of experts model with 2.8 trillion total parameters, 104 billion active per token, and a context window of 1,048,576 tokens. It takes text and images. Those are vendor reported specs, not something we measured.
The Kimi product family at a glance
| Product | What it is | Best for |
|---|---|---|
| Kimi (web and mobile) | The chat assistant at kimi.ai, plus apps for iOS, Android and HarmonyOS | Everyday questions, research, writing, slides and documents |
| Kimi Work | A desktop app for Windows and Apple silicon Macs (version 3.1.0 or later for K3.1 Pro, per Moonshot) | Longer work sessions on your computer |
| Kimi Code | Moonshot's coding tool; you pick K3.1 Pro with the /model command | Developers who want an agent that writes and fixes code |
| Kimi API | Pay as you go access on the Kimi API Platform with OpenAI and Anthropic compatible endpoints | Building Kimi into your own app or agent |
| Open weights | The full K3.1 Pro weights on Hugging Face, under the Kimi K3.1 Pro License | Teams with serious GPU infrastructure that need to run the model themselves |
| Kimi Business and Enterprise | Organization accounts with member management; admins enable K3.1 Pro, which Moonshot says was off by default at launch | Companies rolling Kimi out to a team |
If you're just curious, start with the web app. If you're a developer, the real choice is between the API and the open weights.
Using Kimi through the API
The API is the fastest way for a developer to try Kimi in a real project. You create a key on the Kimi API Platform and call kimi-k3.1-pro or kimi-k3.1-pro-highspeed. Moonshot says the API is OpenAI and Anthropic compatible, so many existing clients work with a base URL change, and that Zero Data Retention is available for API customers.
Kimi's official pricing page lists these rates per million tokens, in US dollars and before taxes (checked November 11, 2026):
| Model | Input | Cached input | Output |
|---|---|---|---|
| kimi-k3.1-pro | $2.00 | $0.20 | $10.00 |
| kimi-k3.1-pro-highspeed | $4.00 | $0.20 | $20.00 |
| kimi-k3 | $3.00 | $0.30 | $15.00 |
The same page says API billing is pay as you go with no subscription, and that it's separate from the Kimi membership and Kimi Code plans. Batch jobs on K3.1 Pro and K3 are billed at 60% of the standard price. Prices change, so check the page the day you budget.
If you're on an older model, note this: Kimi's pricing page says K2 series models stopped accepting newly registered users on October 31, 2026, and that Moonshot plans to retire the K2 series by the end of December 2026. New projects should start on K3 or K3.1 Pro.
Running Kimi yourself: open weights
The K3.1 Pro weights are public on Hugging Face. That's real, and it's rare for a model this size. But open weights aren't the same as "runs on your laptop."
- Hardware. A model with 2.8 trillion parameters needs a multi GPU, datacenter class setup even with Moonshot's MXFP4 quantization. This is a path for infrastructure teams, not for a weekend experiment.
- Inference engines. The model card recommends vLLM, SGLang and TokenSpeed, and links recipes for each.
- License. The weights ship under the Kimi K3.1 Pro License, not a standard open source license. It's permissive for most uses, but it adds conditions. As of November 11, 2026, it says a company running a "Model as a Service" business with more than $20 million in revenue over 12 months needs a separate agreement with Moonshot, and that products above 100 million monthly active users or $20 million in monthly revenue must show "Kimi K3.1-Pro" in their interface. Internal use is exempt. Read the full file yourself, and talk to a lawyer if your product is anywhere near those lines. This isn't legal advice.
How good is it, really?
Moonshot's model card compares K3.1 Pro with leading proprietary models across dozens of benchmarks. For example, it lists a Terminal-Bench 4.0 score of 65.7, run with the Claude Code harness. Treat every one of those numbers as vendor reported: Moonshot chose the benchmarks, the settings and the comparison models, and some benchmarks in the table are internal to Moonshot.
The honest takeaway is simpler. K3.1 Pro is a serious open weight option for coding and agent work, and the only way to know if it fits your task is to try it on your own prompts. If you want to see how it compares with other coding agents in practice, read the Kimi K3.1 Pro section of our guide to the best AI coding agents.
Which Kimi should you use?
- You want a smart assistant for daily work. Use the web or mobile app, or Kimi Work on your desktop.
- You write code and want an agent. Try Kimi Code, or plug the API into a coding tool you already use.
- You're building a product. Start with the API. Move to self hosting only if you have a clear reason, like data residency, and the hardware to back it.
- You run AI for a company. Look at Kimi Business, then read the license before any self hosted deployment.
Keep learning
- Comparing model families? Read what Mistral AI is, another company that publishes open weights.
- More tools explained in plain English live in our AI tools hub.
A next step, no rush
If Kimi has you thinking about building your own agents on top of models like this, take a look at AI Flex, 4Geeks' self paced way to learn applied AI with guided paths and mentorship. Browse it whenever it suits you.
Sources, checked November 11, 2026: Moonshot AI, Kimi K3.1 Pro announcement, Kimi K3.1 Pro model card, Kimi K3.1 Pro License, Kimi API pricing.