MOONSHOT AI · SERVERLESS AND DEDICATED · COMING SOON

kimi-k3

Moonshot's flagship reasoning model with a 1,000,000-token context window, served streamed byte for byte.

1M

TOKEN CONTEXT

$2.10

INPUT, PER MILLION

$10.50

OUTPUT, PER MILLION

Streamed

BYTE FOR BYTE, NO BUFFER

Overview

kimi-k3 is Moonshot AI's flagship open-weight reasoning model, and the largest-context model in the Nyx catalog. The full 1,000,000-token window arrives on serverless with no separate SKU and no long-context surcharge: tokens are the unit whether the prompt is four hundred tokens or nine hundred thousand. Capacity is being provisioned now.

The window is what makes it the default choice for agent workloads: long sessions that accumulate tool output, file contents, and conversation history without summarizing away the details. Prefill runs in slices interleaved with decode, so a million-token prompt cannot park in front of a short one.

Serving on Nyx

Tokens will stream byte for byte the moment the model produces them, with nothing buffering chunks between the engine and your socket. Past declared capacity Nyx returns 429 immediately with a Retry-After header instead of queueing your request into a timeout.

The engine's prefix cache carries across turns, which removes 31 percent of prefill work on agent traces. Dedicated pods are available for pinned capacity with guaranteed throughput at a flat hourly price.

Provider A168
Provider B121

OUTPUT PRICE · $/M · LOWER IS BETTER

Nyx$10.50
Provider A$14.25
Provider B$16.80

Independent harness · run weekly · streamed

Nyx$2.10$10.50214
Provider A$2.85$14.25168
Provider B$3.36$16.80121
# OpenAI-compatible: swap the base URL
curl https://api.nyxprovider.com/v1/chat/completions \
  -H "Authorization: Bearer $NYX_API_KEY" \
  -d '{
    "model": "kimi-k3",
    "stream": true,
    "messages": [{"role": "user",
      "content": "Read the attached repo and plan the migration."}]
  }'
CONTEXT1,000,000 tokens
INPUT$2.10 / M tokens
OUTPUT$10.50 / M tokens
STREAMINGByte for byte, no buffer
PARAMETERS1T total · 32B active
MODALITYText in · text out
LICENSEModified MIT
DEPLOYMENTServerless · Dedicated · coming soon
APIOpenAI-compatible
CATALOG

All models in the library