Moonshot / API REFERENCE

kimi-k3 API pricing & integration

Moonshot’s flagship for long materials, visual understanding and demanding coding. Evaluate conversation integrity, output quality and cache costs.

Estimate a request →Integration guide →

Catalog identifier: kimi-k3
Channel protocols: Chat Completions
Availability: Verify account channels
Snapshot date: Oct 15, 2026

Platform price snapshot

Rates include the selected group multiplier. All token rates below are USD per million tokens; request rates are separate. A pricing boundary is not the model’s maximum context window.

Prompt tier (tokens)	Input	Output	Cache read	Cache write
≥ 0	$11	$55	$1.1	—

Group: 稳定k3. Cache duration, additional modalities, time-of-use rules and channel restrictions may affect the final bill. Check the interactive market and account terms.

Vendor reference pricing

No verified equivalent vendor quote is recorded. We do not infer a vendor rate from a similar model name.

Tasks to evaluate

Provide an error screenshot and relevant code. Ask for visible evidence, a diagnosis and verification steps. Check against known answers and retain complete messages.

Limits and cost considerations

Thinking is always on, with low/high/max reasoning_effort. Preserve complete assistant messages for multi-turn/tools. Official image input uses base64 or ms://, not public image URLs. Cache writes are billed separately.

Vendor capability announcement ↗

Source reviewed: Oct 19, 2026

Compare prices and estimate a request

Published price snapshot, USD per million tokens. Account availability and the actual bill must be checked separately. Route names retain their configured labels.

Snapshot checked: 2026-10-15

Base-tier rates are available below without JavaScript. Enable JavaScript to filter or calculate.

Model	Provider	Input / 1M	Output / 1M	Per-request rate
kimi-k3	Moonshot	$11	$55	—

Token estimates are available for reviewed text-model routes and cover uncached text input and output only. It excludes cache writes, tools, other modalities, time-of-use adjustments, retries and processing premiums. A rate tier is not a context limit. Missing prices stay unavailable rather than becoming zero. Selection uses the lowest input rate, not necessarily the lowest total task cost.

A small integration acceptance check

POSIX shell template for a channel offering Chat Completions. Set the account-provided base URL (normally including /v1) and API key in your terminal. Confirm this exact identifier is enabled. Running it makes a billable request; this page does not send it.

curl "$YOUJIE_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $YOUJIE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"kimi-k3","messages":[{"role":"user","content":"Return a JSON object with the sum of 17 and 25 under the key total. No other text."}],"stream":false}'

Acceptance target: parse the final message as JSON and verify total equals 42. This is a known mathematical target, not a recorded model response. Also inspect HTTP status, response structure and actual usage. No tool calling or strict-schema feature is assumed.

Inspect errors, credentials and protocol paths →

Integration checks

Confirm the exact identifier and enabled group in your account.
Choose a supported protocol, then keep the API key on your server.
Run a small representative task and inspect usage, errors and output quality.

Request examples and troubleshooting →