K3 bills three ways — $0.30 cache-hit input, $3 fresh input, $15 output, per 1M tokens. Your cache-hit rate decides everything, so this calculator makes it a first-class input.
Unofficial tool. Not affiliated with Moonshot AI or Kimi. Prices and scores are gathered from public sources and may lag official changes.
K3 monthly total
$942
Input spend
$342
Output spend
$600
Saved by caching
$1,458
≈ $0.19 per call · blended $1.47 per 1M tokens at 90% cache hits
All models are billed with the same cache-hit rate for a fair comparison (each vendor discounts cached input ~90%). Remember that K3 always reasons: its reasoning tokens bill as output, so on short tasks its real output usage runs higher than models with a non-thinking mode.
Flat rates across the full 1M-token context window — no long-context tier.
| Billing line | Price / 1M tokens | Notes |
|---|---|---|
| Input — cache hit | $0.30 | 90% off; repeated system prompts and agent-loop prefixes land here |
| Input — cache miss | $3.00 | Fresh tokens the cache has not seen |
| Output | $15.00 | Includes reasoning tokens — K3 reasons on every call |
Three billing lines: $0.30 per 1M input tokens on cache hits, $3.00 per 1M on cache misses, and $15.00 per 1M output tokens. The rates are flat across the full 1,048,576-token context window — there is no long-context surcharge tier.
Input tokens whose prefix Moonshot has already processed recently — typically your system prompt, tool definitions, and the unchanged head of a long conversation or agent loop. Moonshot reports cache-hit rates above 90% on coding workloads, which pulls real input cost toward the $0.30 floor.
K3 reasons on every call — reasoning is always on and reasoning_effort only accepts max. Reasoning tokens bill as output at $15 per 1M, so verbose reasoning on simple calls is the main way K3 bills surprise you. Route trivial calls to Kimi K2.7 Code or K2.6 instead.
It uses list prices published in July 2026 and lets you set your own cache-hit rate. It excludes provider-side variations (OpenRouter margins, volume discounts, tool-call overhead), so treat results as planning estimates, not invoices. This is an unofficial tool, not affiliated with Moonshot AI.
At list price, K3's $3/$15 undercuts Claude Opus 4.8 ($5/$25) and GPT-5.6 Sol ($5/$30), and matches Claude Sonnet 5's list price — though Sonnet 5 runs at an introductory $2/$10 through August 2026. But K3 always reasons, so on short tasks its real output bill can exceed a model that answers directly. The calculator lets you test your own ratio.
Yes — sign in to NottoAI and chat with Kimi K3 free, no Moonshot key or card required. That is usually the fastest way to sanity-check quality before wiring up the API.
Kimi K3 Toolkit
Chat with Kimi K3 free on NottoAI — no Moonshot key, no card — and sanity-check quality before wiring up the API.
No credit card required · 100 free credits included