Meet Kimi K3, our most capable model.
Kimi K3 is a 2.8-trillion-parameter model with native vision and a 1M-token context window, built for long-horizon coding, knowledge work, and reasoning. It can sustain long engineering sessions with minimal oversight, navigate massive repositories, and take on work that used to be out of reach.
-
Model ID: kimi-k3
-
Pricing: $0.30/MTok cached input · $3.00/MTok uncached input · $15.00/MTok output — 90%+ cache hit rate in coding, so most input costs $0.30.
-
Get started: Kimi K3 - Kimi API Platform
Use the top-level reasoning_effort field to choose how hard K3 thinks: low for the fastest responses, high for a balance of speed and quality, max for the hardest tasks.
Two quick notes: pass the full assistant message back in multi-turn conversations, and don’t switch to K3 mid-session from another model. K3 is also proactive by nature — if you need tighter boundaries, set them explicitly in your system prompt or AGENTS.md.