Kimi
Moonshot AI's model for long-running agentic tasks, coding and image understanding.
Available models.
Prices are in USD per 1M tokens, checked on 5 Nov 2026. Prices are set by the provider and change. Official pricing
| Model | What it does | Parameters | Price per 1M tokens (input / output) |
|---|---|---|---|
| Kimi K3.1 Pro | The newest flagship, for long-running agentic tasks, coding and image understanding. | 2.8T parameters, 1M-token context | API $2 / $10Per public sources, check Moonshot's price list. Open weights. |
| Kimi K3 | The previous flagship, with the same context of over a million tokens. | 2.8T parameters, 1M-token context | API $3 / $15Per public sources, check Moonshot's price list. Open weights. |
| Kimi K3.1 Pro Highspeed | The same capabilities as K3.1 Pro, with output several times faster according to the vendor. | Same model as K3.1 Pro | API $4 / $20Twice the price of K3.1 Pro, check the price list. |
| Kimi K2.7 Code | A coding-focused variant that also understands images. | 1T parameters, 256K-token context | See Moonshot's price listThe K2 line is being retired by the end of December 2026 and is closed to new users from 31 Oct 2026. Open weights. |
| Kimi K2.6 | An older K2-generation model for text and images. | 1T parameters, 256K-token context | See Moonshot's price listThe K2 line is being retired by the end of December 2026. Open weights. |
| Kimi-VL A3B Thinking | A small image-and-text model with reasoning, runs on a single card. | 16B parameters | Free licenceOpen weights. Your cost is the hardware. |
What it is good for.
A good fit for long-running agentic tasks, coding and working with whole repositories and documents thanks to a context of over a million tokens. It handles text and images. The weights are open, so we can run it on your side too.
What to use it with.
Kimi Code CLIOfficial toolMoonshot AI's official open-source terminal agent.We also connect it to
OpenCode
Hermes
OpenClawBERCIK CodeWhere it runs.
In the cloud
AvailableKimi has its own API run by Moonshot AI. Your data goes to the provider, so for sensitive data we recommend running it on your side.
On your side
AvailableThe weights are open, but the main models have trillions of parameters and need a server cluster. The smaller Kimi-VL runs on a single card. We check the licence.
Want to work with Kimi?
Tell us what you need. We will recommend a version and hardware, or another model if it fits better.
Let's discuss what comes next for your business.
Write a few lines about what you do. We reply the same day and book twenty minutes on the phone.
Tap a photo to see the contact.
Filip Berčík
Websites, SEO, AEO, AI infrastructure project manager, 3D modeling, software development
774 915 200 aetheron@aetheron.czPrague