Kimi K3 Explained: Moonshot AI's Biggest Open-Source Model
What Kimi K3 Is, Who Built It, and What It Means for Your Business
Kimi K3 is a new AI model from Beijing-based startup Moonshot AI. The company calls it the world's largest open-source model, with 2.8 trillion parameters.
This guide breaks down Kimi K3 in plain English for business readers. We'll look at what it is, who built it, and what its open-source approach could mean for your business.
We'll also take an honest look at the risks U.S. and regulated companies should consider before adopting a model built in China.
What Is Kimi K3?
Kimi K3 is a large AI model released by Moonshot AI in July 2026.
Moonshot bills it as the world's biggest open-source model, with 2.8 trillion parameters.
Moonshot has released the full model weights under the Kimi K3 License, so anyone can download and adapt them within that license.
Moonshot also claims Kimi K3 outperforms some cutting-edge U.S. systems. No independent third-party benchmarks exist yet at release, so treat that claim with care.
Access Options and Token Pricing
You can use Kimi K3 through Moonshot AI applications, through the cloud application programming interface (API), or by self-hosting the open weights. App access is available inside Kimi.ai, Kimi Work, Kimi Enterprise, the Kimi Agents mobile app, and Kimi Code by using the /model command. Self-hosting is the main data-residency control. You run the 2.8 trillion parameter model yourself under the Kimi K3 License, and our walkthrough on how to run Kimi K3 locally explains how.
Developers connecting through the Kimi K3 API guide use the model ID kimi-k3 with OpenAI-compatible Chat Completions. Access to the model requires an account top-up of at least $1. The model supports a 1 million token context window, natively processes text, images, and video, and keeps thinking always on with reasoning_effort settings for low, high, or the default max.
Under the published Kimi K3 pricing schedule, Moonshot AI bills on a flat pay-as-you-go basis with no tiers by context length. Usage costs $3 per 1 million input tokens, $15 per 1 million output tokens, and $0.30 per 1 million cache-hit input tokens with automatic context caching.
Who Is Moonshot AI?
Moonshot AI is a Beijing-based Chinese startup and the company behind Kimi K3.
It is best known for Kimi, its long-context AI assistant and chatbot.
Its prior open model, Kimi K2, used a Mixture-of-Experts design. Kimi K3 is the successor to that model.
Moonshot is one of several Chinese labs, alongside DeepSeek and Qwen, pushing powerful open models into the global market.
What Does 'Open Source' Mean for Kimi K3?
Open source here means Moonshot has released the full model weights for anyone to download, run, and adapt under the Kimi K3 License.
Moonshot publishes the Kimi K3 weights and code under the Kimi K3 License.
For a business, open weights unlock two things. You can self-host the model on your own hardware, and you can fine-tune it for your own tasks.
Verify the exact license terms on Moonshot's official release before you build on it. Open weights and a fully permissive license are not always the same thing.
- Self-host on your own servers or private cloud
- Fine-tune the model on your own data
- Avoid per-call fees charged by hosted APIs
- Keep sensitive data inside your own environment
What Do '2.8 Trillion Parameters' Mean for a Buyer?
Parameters are the internal settings a model learns during training, and 2.8 trillion is a very large count.
A bigger parameter count signals scale, but scale does not guarantee better quality for your specific tasks.
Large models also need serious hardware to self-host. Running Kimi K3 in-house would demand significant GPU capacity and cost.
Kimi K3 is a Mixture-of-Experts model like Kimi K2, so it activates only part of its parameters per request. Moonshot's model page lists 104B activated parameters out of 2.8T and a 1-million-token context window.
How Does Kimi K3 Compare at a Glance?
Kimi K3 sits among Chinese open models like DeepSeek and Qwen, and competes for attention with U.S. frontier systems like ChatGPT.
The table below is a plain-English glance, not a benchmark. Moonshot's performance claims are not yet independently verified.
| Model | Who builds it | Open source? | Data residency for a U.S. buyer |
|---|---|---|---|
| Kimi K3 | Moonshot AI (China) | Yes, open weights (Kimi K3 License) | China-built; self-hosting is the main control |
| DeepSeek-V3 | DeepSeek (China) | Yes, open weights | China-built; self-hosting is the main control |
| ChatGPT | OpenAI (U.S.) | No, closed API | U.S. vendor with enterprise data terms |
Chinese open models can be strong and low-cost. U.S. frontier models are closed but come with familiar enterprise contracts and data terms.
What Should a Regulated U.S. Business Consider First?
A regulated U.S. business should weigh data residency, self-hosting, and compliance before adopting Kimi K3.
Sending data to a China-hosted API raises real data-residency and compliance questions for legal, healthcare, and finance firms.
Self-hosting the open weights is the main way to keep sensitive data inside your own environment. That mitigates residency risk but adds hardware and staffing cost.
Because Kimi K3 is new, plan a careful evaluation. Test it on a non-sensitive workload first, and confirm the license and security posture before any production use.
- Data residency: where does your data go if you use a hosted Kimi API?
- Self-hosting: do you have the GPU capacity and staff to run the weights privately?
- Compliance: how does use map to HIPAA, GLBA, or the EU AI Act for your firm?
- Verification: no independent benchmarks exist yet, so validate quality yourself
Should Your Business Use Kimi K3?
Kimi K3 may be worth a pilot if you want a low-cost, self-hostable model and can validate its quality yourself.
It is a weaker fit if you need vendor accountability, signed enterprise data terms, and proven benchmarks on day one.
The safest path for a regulated firm is to self-host the open weights, keep sensitive data in-house, and benchmark it against a model you already trust.
If you are unsure whether Kimi K3 fits your compliance needs, a short vendor-neutral review can save you a costly wrong turn.
What you need to run Kimi K3 yourself
Kimi K3 is a frontier-scale Mixture-of-Experts model, so "running it yourself" is a real infrastructure decision — not something a single laptop or gaming GPU can do. Match the path below to how seriously you need to self-host. For most teams the API or rented GPUs are the right answer; buying hardware only pays off at steady, high volume or when your data can never leave your walls.
| Path | What it is | Best for | Get started |
|---|---|---|---|
| Call the hosted API | Use Kimi K3 as a pay-per-token API — zero hardware | Most teams; evaluating before committing | OpenRouter |
| Rent GPUs by the hour | Spin up H100 / A100 nodes on demand, tear them down after | Self-hosting without capital outlay; bursty workloads | RunPod |
| Local on unified memory | A single workstation with enough unified memory to hold a 4-bit quant | One powerful on-prem box; privacy-first solo/SMB use | Apple Mac Studio (M3 Ultra, 512GB) |
| Local on workstation GPUs | Multiple 48GB professional cards for MoE offload / tensor parallelism | Power users and small clusters that want cards they own | NVIDIA RTX 6000 Ada (48GB) |
Once Kimi K3 is running, the fastest way to put it to work day to day is inside Cursor — point it at the model through OpenRouter as a custom model. And if you would rather run a model on one affordable box, see Best mini PCs for local AI and Local AI hardware calculator.
Frequently Asked Questions
- Kimi K3 is a new AI model from Moonshot AI, a Beijing startup. Moonshot bills it as the world's biggest open-source model, with 2.8 trillion parameters.
- Moonshot AI makes Kimi K3. Moonshot is a Beijing-based Chinese startup best known for its long-context Kimi assistant and its prior open model, Kimi K2.
- Yes. Moonshot has released the Kimi K3 weights under the Kimi K3 License, so people can download and adapt them. Read the license terms in Moonshot's Kimi K3 repository on GitHub before you build on it.
- Moonshot says Kimi K3 has 2.8 trillion parameters, which it bills as the world's biggest open-source model. A big count signals scale but does not guarantee better results.
- Moonshot claims Kimi K3 outperforms some cutting-edge U.S. systems, but no independent benchmarks exist yet at release. Test it against models you already trust before you decide.
- Using a China-hosted Kimi K3 API raises data-residency and compliance concerns. Self-hosting the open weights keeps sensitive data in your own environment and is the main mitigation.
- Moonshot's API lists Kimi K3 at $3 per million input tokens and $15 per million output tokens, with cache-hit input at $0.30 per million. Self-hosting the open weights shifts cost to your own hardware.
The complete AI playbook for your team
Cut your AI bill with Chinese open-weight models — without the risk: Safety, pricing and savings for Kimi K3, DeepSeek, Qwen and z.ai GLM — the four-vendor comparison for owners and IT leads.
Get the guide — $59 (reg. $89)