Kimi K3.
Kimi K3 is Moonshot AI's open-weight multimodal agentic model, 2.8T parameters with 104B active, taking text, images and video with a 1M-token context, under the Kimi K3 licence.
8× AMD MI325 · $24.57/hr.
Most-aggressive quantisation we have a working recommendation for. Lower precision = less VRAM = cheaper hardware, at a small accuracy cost.
Cheapest hosted endpoints.
Variants in the Kimi family.
Moonshot's frontier open-weight MoE — 1T total, 32B active.
Moonshot's open-weight reasoning variant — extended chain-of-thought training...
Multimodal agentic variant — adds a vision encoder to the K2 backbone.
Long-horizon coding + autonomous-execution upgrade over K2.5.
Kimi K2.7 Code is a modified-MIT coding agent model, 1T total and 32B active,...
Kimi K3.1 Pro is Moonshot AI's flagship open-weight multimodal reasoning mode...
Kimi K2 0905 is a modified-MIT mixture-of-experts model, 1T total and 32B act...
Cheapest hardware per quantisation.
Each row is one quantisation tier (the same weights compressed differently). Lower precision → lower VRAM → cheaper hardware, at the cost of small accuracy loss. $/h es la tarifa más baja disponible ahora. Una configuración multitarjeta usa una instancia real de ese tamaño cuando se alquila; si no, es la tarifa de una tarjeta por el número de tarjetas.
| Quantisation | Cheapest GPU config | Total VRAM | Live $/hr | tokens/sec | |
|---|---|---|---|---|---|
|
INT4
INT4 — 4-bit integer (~4× VRAM saving)
|
2048 GB |
$24.57/hr
Cyfuture AI · bajo demanda
|
— | Compare → |
What it costs per month across providers.
Estimate your monthly bill for Kimi K3 across every host that publishes per-token pricing. Slide your token volumes; the chart + table re-rank cheapest-first.
El proveedor más barato, a la izquierda.
Coste mensual total — tokens de entrada y salida juntos.
Desglose de la factura.
| Proveedor | Total mensual | |
|---|---|---|
| $33.92 | Registrarse ↗ | |
| $57.0 | Registrarse ↗ | |
| $60.0 | Registrarse ↗ |
Alquila la GPU en lugar de pagar por token.
Para un modelo de pesos abiertos como Kimi K3, puedes alquilar una GPU y servir la inferencia tú mismo. La cuenta: alquiler de GPU más barato × 730 horas/mes. La electricidad ya está incluida en el precio por hora del alquiler.
Supone que la GPU funciona 24/7 con ~85 % de uso. Si tu tráfico es irregular, pagarás menos por la API y probablemente más por la GPU (las horas ociosas también se cobran). El análisis del punto de equilibrio está en la herramienta de punto de equilibrio autoalojar vs. API.
About Kimi K3.
Kimi K3 is an open-weight multimodal agentic model from Moonshot AI with 2.8 trillion parameters in total and 104 billion active, using a mixture-of-experts design that routes each token to 16 of 896 experts. It understands text, images and video and has a 1-million-token context window. Moonshot aims it at long coding sessions, research and knowledge work. Weights are released under the Kimi K3 licence, not a standard open-source one.