Moonshot AI · updated 2026-08-29
Run Kimi K3: providers, pricing and flat-rate access
Moonshot's 2026 flagship — a frontier-class model that became a favorite for hard agentic coding, priced accordingly.
Kimi K3 is expensive to run per-token, which is exactly why people search for a subscription: Moonshot's own app tiers ($19–199/mo) cover chat but not heavy API use. For API access, OpenRouter serves it per-token; Standard Compute includes K3 in its flat-rate routed pool, so agents reach it without per-token bills.
Your options, honestly
Moonshot app plans$19–199/moChat/app tiers — not built for high-volume API agent use.
Moonshot / OpenRouter APIPay-per-tokenFull API access; K3 is a premium-priced model, so agentic sessions get expensive fast.
Standard Compute$89+/mo flatK3 in the routed pool for hard reasoning steps — the frontier step without the frontier bill. Routing decides when it's used.
Self-host—Not practical: K3 weights aren't openly available at frontier quality.
the flat-rate trade-off
K3 is the model where per-token pain is sharpest — a heavy K3-only agent can pass $30/day per-token. Flat-rate makes sense precisely because routing reserves K3 for the steps that need it.
Running it in your agent
OpenAI-compatible agents reach K3 via OpenRouter (pinned) or via Standard Compute's routed 'standardcompute' model (automatic). Claude Code users route through /v1/messages endpoints.
FAQ
Why is my Kimi K3 bill so high?
K3 is priced at the frontier tier, and agents re-send growing context every step — so long sessions multiply an already-premium rate. Reserve K3 for hard steps, or move to a plan where the price is fixed.
Can I get Kimi K3 on a flat monthly price?
Moonshot's own flat tiers are app-focused. For API/agent use, multi-model flat plans that include K3 in routing are the practical flat-rate route.
More: how flat-rate LLM APIs work · per-agent guides for OpenCode, Claude Code and Cline · model comparisons