Kimi K3 Pricing: API Costs, Cache Economics, and How It Compares

Kimi K3's launch pricing marks a strategic shift for Moonshot AI: instead of undercutting frontier models, K3 is priced in the top tier — and bets that cache economics make it cheaper in practice than the sticker price suggests.

Official API Pricing

Item	Price (per 1M tokens)	China price
Input (cache miss)	$3.00	¥20
Input (cache hit)	$0.30	¥2
Output	$15.00	¥100

Moonshot is also running a top-up bonus promotion of up to 30% at launch.

The Cache Is the Real Story

K3 is served on Moonshot's Mooncake disaggregated inference architecture, and in real coding workloads the prompt-cache hit rate exceeds 90%. That matters because cache-hit input costs 10× less than uncached input:

• A typical agentic coding loop re-sends a large, mostly unchanged prefix every turn
• With a 90%+ hit rate, effective input cost is roughly a quarter of list price (closer to ~$0.57/1M blended than $3.00)
• Output ($15/1M) is uncacheable, so it dominates long-generation workloads

How K3 Compares

• Claude Sonnet 5 — essentially identical sticker price ($3/$15)
• Claude Opus 4.8 — K3 lands roughly 40% cheaper
• GPT-5.6 — K3 undercuts it
• Kimi K2.7 Code — K3's output price is about 4× the previous generation; the quality jump is the justification
• DeepSeek V4 Pro — far cheaper (single-digit yuan per million tokens), but not in the same capability class

The takeaway: K3 is no longer the "near-frontier performance at a fraction of the price" play that earlier Kimi models were. Moonshot is pricing K3 as a genuine frontier model and letting the cache architecture deliver the discount where it counts — iterative, prefix-heavy workloads like coding agents.

Membership and Context Tiers

API access aside, K3 in Kimi's own products is gated by plan:

• Moderato and above: K3 access with 256K context
• Allegretto and above: full 1M token context (set 1048576 as the context window in third-party tools)
• Andante: K3 not available

For the full model specs, see our Kimi K3 launch coverage. Official pricing lives at platform.kimi.com/docs/pricing.

Prices verified July 17, 2026. Check the official docs for the latest rates.