The open-weight flagship that matches closed rivals on modality — images, video, a 1M window — at half their price. Real per-million-token rates, a workload calculator, and the head-to-head comparisons.
| MODEL | INPUT $/1M | CACHED IN | OUTPUT $/1M | BLENDED | CONTEXT | YOUR COST |
|---|---|---|---|---|---|---|
Kimi K3 kimi-k3 · flagship | $3.00 | $0.30 | $15.00 | $6.00 | 1.05M | $10.50 |
Kimi K2.7 Code kimi-k2.7-code · coding | $0.95 | $0.19 | $4.00 | $1.71 | 262K | $2.95 |
Kimi K2.6 kimi-k2.6 | $0.95 | $0.16 | $4.00 | $1.71 | 262K | $2.95 |
USD per 1M tokens from Moonshot AI’s official pricing page as of August 31, 2026. Blended assumes a 3:1 input:output ratio. Get keys at the Moonshot AI console.
Kimi K3 is Moonshot’s case that open weights and frontier capability aren’t a trade-off: $3 input / $15 output per 1M tokens for a reasoning model that takes images and video in a 1.05M-token context. That is flagship territory — Sol-class modality — at roughly a third of Claude Opus 5’s blended price.
Below it, K2.7 Code and K2.6 share a $0.95 / $4 price point with 262K context. K2.7 Code is the one tuned for agentic coding — it competes head-on with Grok Build and undercuts Claude Sonnet 5 by more than half on blended cost.
Cache reads bill at a tenth of fresh input ($0.30 on K3), and because the weights are open you can also self-host or route through OpenRouter — the first-party platform at platform.kimi.ai is usually the cheapest managed option.
Kimi K3 costs $3 input / $15 output per 1M tokens; K2.7 Code and K2.6 both cost $0.95 / $4. Cache reads are about a tenth of the fresh input rate. All three ship open weights.
K2.6 at $0.95 input / $4 output per 1M tokens ($1.71/1M blended). If your workload is coding, pick K2.7 Code instead — same price, tuned for agentic editing.
K3 blends to about $6/1M tokens versus $8 for GPT-5.6 Sol and $10 for Claude Opus 5, while matching them on vision input and adding video — capabilities Claude doesn’t price at all. The closed flagships still lead on some hard-reasoning benchmarks; K3 is the value play at the frontier.
Yes — K3 and the K2 line ship open weights, so you can run them on your own hardware or via any inference provider. The rates here are Moonshot’s first-party API prices from platform.kimi.ai.
K2.7 Code at $0.95 / $4 — purpose-tuned for code generation and agentic editing, with 262K context for large repos. It is the standard budget alternative to Claude Sonnet 5 in coding agents.
Also see: OpenAI API pricing · Claude API pricing · Gemini API pricing · Grok API pricing · DeepSeek API pricing · GLM API pricing · MiniMax API pricing · MiMo API pricing · all 58 models
ReqKey gives every key a credit balance — validate, deduct, enforce limits, and watch per-consumer Kimi spend in real time.