"Free" means three different things here, and conflating them is why people get confused. The Kimi apps have free consumer access. The API is paid. The weights are free to download and expensive to run. Here is the honest breakdown of each.
The three answers
| Route | Cost |
|---|---|
| Kimi consumer apps | Free tier available; plan limits change over time |
| Kimi API | Paid — $3.00 input / $15.00 output per million tokens |
| Cached API input | $0.30 per million tokens |
| Open weights | No licence fee; substantial hardware cost |
Which capabilities sit behind which app plan changes over time, so check the app rather than trusting a blog post — including this one.
"Free weights" is not free inference
The weights carry no licence fee, and this is the number most often misread as "free". Running them requires ~1.4 TB of memory across a multi-node GPU cluster, plus the electricity and the people to operate it.
For nearly every organisation, API access at $3.00 per million input tokens is dramatically cheaper than self-hosting — see hardware requirements for why. Free weights are valuable as an exit option and a check on pricing power, not as a way to avoid paying for compute.
How to spend very little on the API
If cost is the concern, the API is cheaper than most people assume:
- A short chat exchange costs under two cents.
- Reviewing a 5,000-line diff costs around 23 cents.
- Cached input at $0.30 makes repeated work against the same context nearly free on the input side.
reasoning_effort: lowcuts the reasoning tokens that bill as output.
For evaluation and personal projects, you will likely spend less on the API in a month than on a single consumer subscription.
More on Kimi K3
Start with the complete Kimi K3 guide for the overview, or go deeper:
Ready to go deeper?
Read the full Kimi K3 guideFrequently Asked Questions
Is Kimi K3 free to use?
The Kimi consumer apps have a free tier, though which capabilities it includes changes over time — check the app. The API is paid at $3.00 per million input tokens and $15.00 per million output. The open weights carry no licence fee but require substantial hardware to run.
Can I use Kimi K3 without paying anything?
Through the Kimi consumer apps, within whatever the current free tier allows. For API access you pay per token, though the amounts are small for individual use — a short exchange costs under two cents.
If the weights are free, why would I pay for the API?
Because free weights are not free inference. Running 2.8 trillion parameters means ~1.4 TB of memory across a multi-node GPU cluster, plus power and staff. For almost everyone, $3/$15 per million tokens is far cheaper than self-hosting.
