Kimi K3 pricing for rag pipeline

Quick start

Call Kimi K3 with the OpenAI Python SDK pointed at api.moonshot.cn:

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.moonshot.cn/v1",
    api_key=os.environ["MOONSHOT_API_KEY"],
)
response = client.chat.completions.create(
    model="kimi-k3",
    max_tokens=1500,
    messages=[{"role": "user", "content": "Your prompt here"}]
)
# Cost: ~$0.0405 per call at standard rates
# Input:  $3.00/MTok
# Output: $15.00/MTok

How Kimi K3 compares on cost

Per-call cost for all models scoring 50+ on rag pipeline benchmarks. Hover for details.

Cheapest viable option: GPT 5.6 Luna at $0.003000/call (93% less, 85/100 quality).

What drives rag pipeline cost

Retrieval-augmented calls are dominated by the retrieved passages, not the question. Retrieving eight chunks instead of three roughly triples input cost for an answer that is often no better, so retrieval tuning is cost tuning. The system prompt and instructions are identical on every call, which makes cache reads the first thing to enable.

For Kimi K3 specifically, that means output tokens cost 5.0x more than input, at $15.00 vs $3.00 per MTok. At this use case’s 6,000/1,500 token split, output accounts for 56% of the $0.0405 per-call cost.

Kimi K3 quality benchmarks for rag pipeline

Benchmark data for rag pipeline is pending for Kimi K3. Test against your own evaluation suite before moving to production.

Benchmark data is not yet available for Kimi K3. We recommend testing against your own evaluation criteria.

About Kimi K3

Prices from platform.kimi.ai (read 2026-10-11). Cache hit: $0.30/MTok input. Kimi K3 is a 2.8T-parameter open-weight MoE model. No batch discount published yet. Not yet benchmarked, so never recommended.

Other models from Moonshot (Kimi)

You can compare Kimi K3 head-to-head against any other model on our model comparison pages, or track how pricing has changed over time on the pricing history tracker.

Frequently asked questions