29 models.
One OpenAI-compatible endpoint.
Point your existing OpenAI client at KI and reach a curated catalog of capable models. Your subscription covers everyday inference, and your prompts and responses are never logged or stored.
- Current models
- 29
- Largest context
- 1.1M tokens
- Plans from
- $15/mo
The catalog
Cost is relative to a typical model: fewer coins, lighter per call.
- Mistral Small 4 mistralai/mistral-small-2603 260K 0.5x the cost of a typical model
- MiMo-V2.5 xiaomi/mimo-v2.5 1M 0.3x the cost of a typical model
- MiMo-V2.5-Pro xiaomi/mimo-v2.5-pro 1M 0.9x the cost of a typical model
- MiMo-V2.6-Flash xiaomi/mimo-v2.6-flash 1M 0.3x the cost of a typical model
- MiMo-V2.6-Pro xiaomi/mimo-v2.6-pro 1M 0.9x the cost of a typical model
- Hermes 4 405B nousresearch/hermes-4-405b 131K 2.7x the cost of a typical model
- Kimi K3 moonshotai/kimi-k3 1M 12x the cost of a typical model
- MiniMax M3 minimax/minimax-m3 1M 1x the cost of a typical model
Four steps to your first response.
Keep your OpenAI client. Change the base URL, keep the rest.
01 Create an account
Sign up in seconds. No card required to start.
02 Swap the base URL
Point your OpenAI client at the KI endpoint and drop in your key.
03 Pick a model
Choose any model by name. Switch with a string, not a rewrite.
04 Monitor usage
Track requests and remaining compute from one dashboard.
Simple monthly plans. No token math.
Both plans include every model. The only difference is the size of the weekly compute allowance.
See pricing1x weekly allowance
A weekly compute allowance for everyday inference.
2x weekly allowance
Twice the allowance for heavier, high-volume workloads.
No content logging. No training.
Every request is proxied in real time. KI never logs your prompts or trains on your data. Providers may briefly store a request to process it, never to train.