Moonshot: kimi-k3

Kimi‑K3 is Moonshot AI’s new flagship open‑weight MoE large model with 2.8 trillion total parameters. It features a 1‑million‑token ultra‑long context window and native vision multimodal input. Built on KDA hybrid linear attention architecture, it excels at large‑codebase processing, complex knowledge analysis and deep reasoning. Full weights are open‑sourced, supporting function‑call and structured output.

Modetext → text

Input price$3 per M tokens

Output price$15 per M tokens

Context length

Weekly usage34.917K tokens

Listed at

Moonshot

Latency—

Throughput—

Upload rate—

Context length—

Max output—

Input price$3per 1M tokens

Output price$15per 1M tokens

Cache read$0.8999999999999999per 1M tokens

Cache write$0per 1M tokens

Performance

Avg TPS—

Avg latency—

Avg success rate—

TPS—

TTFT—

Latency—

Success rate—

Success rate

Speed

Code Example

curl ${API_BASE_URL}/v1/chat/completions \
  -H "Authorization: Bearer <YOUR_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k3",
    "messages": [
      {
        "role": "user",
        "content": "Hello"
      }
    ]
  }'

Replace <YOUR_API_KEY> with the API key generated on your token management page.

Authentication

All requests must include the Authorization: Bearer <TOKEN> header.

Endpoint Path

POST

openai

Supported Parameters

ParameterTypeDefaultDescription

Rate Limits

GroupRPMTPMRPD

RPM = requests per minute, TPM = tokens per minute, RPD = requests per day. Limits apply by token group.

Pricing

GroupBilling typePrice summary

MoonShot0%

Per call

Input price3per 1M tokens

Completion price15per 1M tokens

Cache read price0.8999999999999999per 1M tokens

Cache write price0per 1M tokens