Kimi K3 has three pricing routes: a free chat tier, four paid memberships from $19 to $199/mo, and a pay-per-token API at $3/$15 per million. On top of that, the weights are open, so you can self-host if you have the hardware. The right choice depends on whether you're chatting, building or running agents at scale — here's the breakdown.
Kimi pricing at a glance
| Route | Price | Best for |
|---|---|---|
| Adagio (free) | $0 | Casual chat, trying the model |
| Moderato | $19/mo | Individuals wanting Kimi Code + Deep Research |
| Allegretto | $39/mo | Regular agentic and coding use |
| Allegro | $99/mo | Heavy Agent Swarm + Professional Data |
| Vivace | $199/mo | Maximum concurrency and quotas |
| API | $3 / $15 per 1M | Developers, apps, high-volume agents |
The membership tiers
Memberships are named after musical tempos and scale up quota and agent capabilities rather than raw model access — every paid tier runs K3. Moderato ($19/mo) is the entry point, unlocking Kimi Code (the coding agent) at entry credit levels, Deep Research, and the Slides and Websites generators.
Allegretto ($39/mo) and Allegro ($99/mo) raise Kimi Code credits and open up Agent Swarm — Moonshot's parallel-agent mode that runs up to 300 subagents — plus Kimi Claw cloud deployment and growing Professional Data quotas. Vivace ($199/mo) is the top tier, for maximum concurrent tasks and the largest allocations.
One thing to watch: the paid plans run on weekly usage quotas for Agent Swarm and Kimi Code credits. Run heavy agentic sessions on Moderato and you can exhaust the weekly allocation before it resets — which is the main reason to step up a tier. Annual billing saves roughly $48 / $96 / $240 / $480 across the four plans.
Free tier (Adagio)
The Adagio free plan lets you chat at kimi.com with no credit card. It uses a lighter model variant and imposes daily quotas, but it's a real way to try K3's reasoning and long-context handling before paying. For casual questions and light use it's genuinely usable, not a crippled demo.
What it doesn't give you is meaningful agentic capacity — Kimi Code and Agent Swarm are where the paid tiers earn their keep. If you're just kicking the tires or using Kimi as a second-opinion assistant, Adagio is plenty.
API pricing
For developers, the API is the headline. K3 costs $3 per million input tokens and $15 per million output tokens, with cached input at $0.30 per million. That output rate matches Claude's Sonnet 5 and undercuts the top closed flagships (Claude Opus 4.8 runs $5/$25) while delivering elite coding quality.
The cached-input rate is the quietly important number. Agentic loops re-read a large context on every step, so at $0.30 per million cached tokens, running an agent over a big codebase costs a fraction of what it would on a model without cheap caching. For token-heavy, long-context workloads, K3 is one of the best value options on the market. Unlike the app memberships, you pay only for what you consume.
Self-hosting the open weights
Because K3 is open-weight (release promised by July 27, 2026), you can run it on your own infrastructure for the cost of the hardware. The catch is the hardware: this is a 2.8-trillion-parameter model, and even with only 16 experts active per token, serving it needs a serious multi-GPU setup.
Self-hosting makes sense in two cases: very high sustained volume where per-token API costs would dwarf infrastructure spend, or strict data-residency requirements where prompts can't leave your network. For everyone else, the API is cheaper and far less operational hassle.
Which route is worth it?
Chatting? Start free on Adagio; upgrade to Moderato ($19) only when you want Kimi Code and Deep Research regularly. Building an app or running agents? Use the API — $3/$15 per million with $0.30 caching is hard to beat for coding-heavy work. At massive scale or under data-residency rules? Evaluate self-hosting.
For a sense of where K3 lands against the closed competition, our Kimi K3 review covers the benchmarks and the Kimi vs Claude comparison weighs it against the writing-and-coding benchmark. If you also want the closed side's plan-by-plan costs, see our ChatGPT pricing guide.
Frequently Asked Questions
How much does Kimi cost in 2026?
Kimi has a free tier (Adagio) and four paid memberships: Moderato $19/mo, Allegretto $39/mo, Allegro $99/mo and Vivace $199/mo. Separately, the K3 API costs $3 per million input tokens and $15 per million output tokens.
Is Kimi free to use?
Yes. The Adagio free tier lets you chat at kimi.com with no credit card, using a lighter model variant with daily quotas. Paid tiers add higher limits, Kimi Code credits, Agent Swarm and Deep Research.
What is the difference between Kimi membership and the API?
Memberships give you the chat app plus agent features on weekly usage quotas — good for individuals. The API bills per token ($3/$15 per million) and is better for developers building apps or running agentic workloads at volume, where you pay only for what you use.
Which Kimi plan is best value?
For most people the Moderato plan at $19/mo covers Kimi Code, Deep Research and the Slides and Websites tools. Heavy agentic users who need Agent Swarm and larger quotas step up to Allegretto ($39) or Allegro ($99); Vivace ($199) is for maximum concurrency.
Does annual billing save money on Kimi?
Yes. Annual billing saves about $48 on Moderato, $96 on Allegretto, $240 on Allegro and $480 on Vivace compared with paying monthly.
Is it cheaper to self-host Kimi K3?
The open weights are free, but running a 2.8-trillion-parameter model needs substantial GPU capacity, so self-hosting only pays off at high, sustained volume or where data must stay in-house. For most users the API at $3/$15 per million tokens is cheaper and simpler.