Kimi K3.1 Pro Pricing

TL;DR

• Standard kimi-k3.1-pro list price: ¥13 per 1M input tokens and ¥68 per 1M output tokens.
• Highspeed kimi-k3.1-pro-highspeed list price: ¥26 per 1M input and ¥136 per 1M output tokens.
• Cache-hit input costs ¥1.30/MTok; 5-minute cache writes cost ¥13/MTok and 1-hour writes cost ¥26/MTok.
• A workload of 50 requests/day using 2,000 input and 500 output tokens costs about ¥90 per month before caching.

K3.1 Pro price table

Monthly workload examples

These estimates assume 30 days and exclude cache writes, retries, and hidden reasoning tokens. Input is 2,000 tokens per request; output is 500 tokens.

How much prefix caching saves

Suppose a 100,000-token policy document is reused for 1,000 questions. Without caching, 100M input tokens cost about ¥1,300. With a cache hit, the same repeated prefix costs about ¥130 plus a small write fee, roughly a 90% input saving.

Caching works best when the system prompt or document prefix is stable. If every request changes the beginning of the prompt, the cache cannot be reused.

When Highspeed is worth 2×

• Use standard K3.1 Pro for batch analysis, deep research, background code review, and cost-sensitive workloads.
• Use Highspeed when user-facing latency or output throughput directly affects product quality.
• Measure both variants on the same prompt before switching: faster output is not worth 2× if answer quality or tool reliability drops.

Cost-control checklist

• Set max_completion_tokens below your product ceiling.
• Cache stable instructions and long reference documents.
• Log input tokens, output tokens, retries, and tool-loop steps per conversation.
• Route simple classification or formatting tasks to a smaller model.
• Choose a 1-hour cache TTL only when requests are likely to arrive after five minutes.