AI subscription costs have quietly become a real budget line item. Some developers are sitting on $200/month for Claude Max, another $40 for GitHub Copilot, and a separate API key they justified as “just testing.” It adds up faster than it should.
So when Kimi’s API pricing started circulating in developer communities, the reaction was predictable: either “this is too cheap to be real” or “there must be a catch buried somewhere.” Neither is exactly right. Here’s the full picture: what Kimi is free for, what each tier actually unlocks, how the API billing works, and where costs can sneak up on you.
The Most Important Thing to Know First
Kimi subscription plans and Kimi API access are completely separate systems. They are never bundled.
You can hold a paid subscription and still pay per token on the API. You can call the API with no subscription at all. Paying for Moderato does not discount your tokens. Topping up API credits does not unlock subscription features.
Why does this matter? Because most of the pricing confusion around Kimi comes from assuming one unlocks the other. It doesn’t. Keep the two systems mentally separate and everything becomes cleaner.
Subscriptions cover the Kimi consumer interface at kimi.com: Agent Swarm access, Kimi Code credits, Deep Research runs, Kimi Slides, and productivity tools.
API billing covers developer access through api.moonshot.ai, charged per token. Completely independent.
The Free Tier: What You Actually Get
The free tier at kimi.com is real and usable for casual work. No credit card required. You get web search integration, file uploads (documents, PDFs, images), and reasonable daily message quotas.
The free tier runs an older or lighter model variant, not the full K2.6 flagship. For most general tasks, the quality difference isn’t dramatic. For heavy coding sessions, extended agentic runs, or anything requiring Agent Swarm, you’ll hit the limits quickly.
On the API side, there’s no ongoing free tier. The minimum to activate an account is $1. When your cumulative recharges hit $5, you receive a $5 voucher, effectively doubling your first top-up. Vouchers don’t count toward the cumulative recharge total that sets your rate-limit tier.
Subscription Plans: The Full Breakdown
Moonshot uses musical tempo names for its subscription tiers. The current lineup per Moonshot’s published pricing page:
| Plan | Price | Core Access |
|---|---|---|
| Adagio | Free | Basic chat, light daily use, limited daily quotas |
| Moderato | $19/month | K2.6 in chat, Deep Research, Kimi Code entry, Slides and Websites tools |
| Allegretto | $39/month | Higher credit and quota allocations than Moderato |
| Allegro | $99/month | Serious Agent Swarm access, more Kimi Code credits, Professional Data features |
| Vivace | $199/month | 300-agent swarm, Kimi Claw (cloud deployment + browser automation), max quotas |
Here’s what actually changes at each step.
Moderato ($19/month) is the entry-level paid plan. You get K2.6 access in the chat interface, Kimi Code entry-level credits, Deep Research, and the Slides and Websites tools. For individual professionals doing AI-assisted work, this is broadly comparable in scope to ChatGPT Plus or Claude Pro at a slightly lower price.
Allegretto ($39/month) scales up credit and quota allocations. Worth it if you’re consistently hitting Moderato’s daily limits without needing the full professional tier features.
Allegro ($99/month) is where Agent Swarm access becomes meaningful. Higher concurrent task limits, more Kimi Code credits, and the beginning of Professional Data features for document-heavy workflows. Kimi Code credits in this tier are enough for sustained daily coding use. Kimi Slides generation credits are also more generous here than at Moderato.
Vivace ($199/month) is the full-capacity tier. 300-parallel-agent swarm access, Kimi Claw (cloud deployment and browser automation), maximum Professional Data quotas, priority access to new model releases. This is the plan for teams running Kimi as serious agentic infrastructure, comparable in position to Cursor Ultra or ChatGPT Pro.
Annual billing typically saves 15 to 25% versus monthly rates, though exact discounts vary by tier and promotion. Check kimi.com for current annual pricing before committing.
To cancel a Kimi AI subscription: Go to your account settings at kimi.com, navigate to Membership or Billing, and select Cancel Plan. Cancellation takes effect at the end of your current billing period. You retain access to paid features until that date. If you’re on annual billing, refund eligibility depends on your region and how far into the billing cycle you are. Moonshot’s support contact is available through the Help section of the site if you run into issues.
All subscription pricing reflects Moonshot’s published rates as of July 2026 and is subject to change. Verify at kimi.com before subscribing.
API Pricing: Developers, Here’s What Matters
The developer API at api.moonshot.ai (international) is token-based and always separate from subscriptions. Mainland China access runs through platform.moonshot.ai with a separate RMB rate card.
Per Moonshot’s platform documentation as of mid-2026, the main model rates are:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Cached Input |
|---|---|---|---|
| Kimi K2.6 (flagship) | ~$0.95 | ~$4.00 | ~$0.16 |
| Kimi K2.5 | ~$0.60 | ~$3.00 | ~$0.10 |
| Kimi K2.7 Code | ~$0.55 to $0.75 | ~$2.50 to $3.50 | Varies |
These figures reflect publicly available API pricing as of July 2026. Pricing fluctuates across model variants, endpoints, and third-party providers (OpenRouter lists Kimi models at different rates). Always verify current rates at platform.moonshot.ai or platform.kimi.ai before building cost models.
A few mechanics worth knowing before you start spending.
Automatic context caching. When the same or similar context appears across requests (system prompts, document text, conversation history), Kimi caches it and bills at the cached rate, roughly 80 to 85% lower than the standard input price. No configuration required. For production applications with consistent context (a fixed system prompt, a codebase context that doesn’t change between sessions), the effective input cost is substantially lower than the headline rate.
Web search tool cost. Enabling the built-in web search tool adds $0.005 per successful search call, plus the returned results are billed as input tokens on the next turn.
Rate limit tiers. Accounts are grouped into tiers (Tier 0 through Tier 5) based on cumulative recharge, from $1 up to $3,000. Higher tiers unlock more concurrent requests and higher RPM. The per-token price is flat across all tiers: spending more lifts your throughput ceiling, not your unit cost. There is no weekly quota on the API the way there is on some subscription features; billing is purely consumption-based.
The subscription plans, by contrast, do operate on weekly usage quotas for Agent Swarm and Kimi Code credits. If you’re on Moderato and run heavy Agent Swarm sessions, you may exhaust your weekly credit allocation before the reset. This is the primary operational difference between subscription credits and API billing for developers doing agentic work at volume.
What These Prices Actually Mean
Developers used to Anthropic or OpenAI pricing will do a double-take. Let’s put the numbers in context.
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| DeepSeek V4 Pro | ~$0.44 | ~$0.87 |
| Kimi K2.5 | ~$0.60 | ~$3.00 |
| Kimi K2.6 | ~$0.95 | ~$4.00 |
| GLM-5.2 (Z.ai) | ~$1.40 | ~$4.40 |
| Claude Sonnet 4.6 | $3.00 | $15.00 |
| Claude Opus 4.8 | $5.00 | $25.00 |
| GPT-5.5 | $5.00 | $30.00 |
Prices sourced from public pricing pages and API documentation as of July 2026.
At 100 million output tokens per month, the monthly API cost comparison looks roughly like:
- DeepSeek V4 Pro: ~$87
- Kimi K2.5: ~$300
- Kimi K2.6: ~$400
- Claude Sonnet 4.6: ~$1,500
- Claude Opus 4.8: ~$2,500
With context caching active on applications with stable context, those Kimi figures drop further. For a SaaS product with a fixed system prompt and consistent codebase context, the effective monthly cost can be substantially lower than the cache-miss headline rate suggests.
The quality trade-off is real. Claude Opus 4.8 remains stronger on several of the hardest software engineering benchmarks. Whether that gap matters for your specific application is something only your own testing will answer.
The Open-Weight Option (The Free Route)
Kimi K2, K2.5, K2.6, and K2.7 Code weights are all available on Hugging Face under a Modified MIT license. Download them, run them on your own infrastructure, pay nothing to Moonshot per token.
Supported frameworks include vLLM, SGLang, KTransformers, and TensorRT-LLM. INT4 quantization reduces the hardware footprint significantly.
The practical constraint: running the full 1T-parameter model at usable speed requires serious GPU infrastructure. Most teams doing this in production run on cloud GPU instances (A100 or H100 class) rather than on-premise hardware. The cloud compute cost replaces the per-token cost. Whether self-hosting is cheaper than the Kimi API depends entirely on your GPU utilization rates and procurement situation.
For regulated industries where data sovereignty is non-negotiable, self-hosting isn’t just a cost calculation. It’s the only viable path.
Subscription vs API: Which Should You Use?
| Use case | Better option |
|---|---|
| General chat, research, Kimi Slides | Subscription (Moderato or above) |
| Kimi Code for daily development | Subscription coding credits + API for heavy sessions |
| Building apps or APIs on top of Kimi | API only (no subscription needed) |
| Running automated pipelines | API only |
| Privacy/data sovereignty required | Open weights, self-hosted |
| Testing before committing | Free tier + minimum $1 API top-up |
Kimi Subscription vs Claude Code vs Cursor
Developers evaluating Kimi Code against Claude Code and Cursor need to compare carefully, because the billing structures are genuinely different.
| Tool | Subscription | Model | API separate? |
|---|---|---|---|
| Kimi Allegro | $99/month | K2.6 + K2.7 Code | Yes, billed per token |
| Claude Code (individual) | ~$100+/month | Claude Sonnet/Opus | Varies by plan |
| Cursor Pro | $20/month | Various | Credits included |
| Cursor Ultra | $200/month | Various | Higher credits |
Kimi Allegro’s API costs, billed separately, will add to the $99 for heavy agentic coding users. Claude Code’s subscription includes a model-call quota. Cursor’s plans include credits toward model usage.
For developers running heavy agentic sessions all day and primarily wanting the best cost-per-task ratio, Kimi K2.6 or K2.7 Code via API is likely the most economical route, especially with context caching. For developers who want a polished integrated experience and are willing to pay more for it, Claude Code’s ecosystem depth still has an edge.
Frequently Asked Questions About Kimi AI Pricing
Is Kimi AI free to use?
Yes. The basic chat interface at kimi.com is free, no credit card required. The free tier uses a lighter model variant with daily quotas. The developer API requires a minimum $1 recharge to activate, with no ongoing free tier.
What does Kimi Moderato include?
Moderato ($19/month, per Moonshot’s published pricing) gives you K2.6 access in the chat interface, entry-level Kimi Code credits, Deep Research, Kimi Slides, and Websites tools. It’s the entry point for paid subscribers.
What is the Vivace plan?
Vivace ($199/month) is Kimi’s highest published tier. It includes 300-parallel-agent swarm access, Kimi Claw (cloud deployment and browser automation), maximum Professional Data quotas, and priority model access.
What does Kimi’s API cost per million tokens?
Per Moonshot’s platform documentation as of mid-2026, Kimi K2.6 costs approximately $0.95 per million input tokens and $4.00 per million output. Kimi K2.5 is approximately $0.60 input and $3.00 output. Context caching drops repeated input to roughly $0.10 to $0.16 per million. Rates vary by provider and endpoint, so verify current figures at platform.moonshot.ai.
Is there a Kimi API free tier?
No permanent free tier. Minimum $1 recharge to activate. A $5 voucher is issued when cumulative recharges hit $5. Some third-party providers (OpenRouter, Cloudflare Workers AI) have historically carried Kimi models on free tiers, but availability changes. Confirm with the provider before depending on it.
How does Kimi context caching work?
Automatically. When repeated or overlapping context is sent across requests, Kimi caches it and bills at the cached input rate, roughly 80 to 85% lower than standard input pricing. No configuration required.
Is annual billing available?
Yes. Annual billing discounts are available and typically save 15 to 25% versus monthly rates depending on tier. Check kimi.com for current annual pricing, as promotional rates shift.
How does Kimi pricing compare to ChatGPT Plus?
ChatGPT Plus is $20/month. Kimi Moderato is $19/month. At the subscription level they’re broadly comparable. On the API, Kimi K2.6 at approximately $0.95 input and $4.00 output is substantially cheaper than GPT-5.5 at $5.00 input and $30.00 output.
What is Kimi Claw?
Kimi Claw is Moonshot’s cloud deployment and browser automation feature, available on Vivace. It enables Kimi agents to interact with websites and web applications in the cloud on your behalf, rather than requiring a local desktop session.
Does Kimi have volume discounts?
Not on per-token pricing. Spending more increases your rate-limit tier (higher RPM, more concurrent requests), but the per-token price is flat. There are no volume discounts on unit cost.
Final Thoughts
Kimi’s pricing is genuinely competitive, and the more you dig into the mechanics, the better the case gets. The subscription tiers are coherently positioned. The API rates sit well below Claude and OpenAI. The automatic context caching means real production costs are often significantly lower than headline rates.
What you’re trading for that: a younger ecosystem, benchmarks that are still primarily vendor-reported on the newest models, and Chinese data routing if you use the hosted API. Self-hosting resolves the last one at the cost of infrastructure overhead.
For developers evaluating Kimi purely on cost efficiency against models with comparable coding benchmarks, the numbers are hard to argue with. The question is always whether the performance on your specific workload justifies the switch.
That’s something a real trial answers better than any pricing table.
Pricing data reflects Moonshot’s publicly available rates as of July 2026. Subscription prices, API rates, and tier features change frequently. Third-party providers may offer different rates. Always verify current pricing at kimi.com or platform.moonshot.ai before making budget decisions.
Curated by Lorphic
Digital intelligence. Clarity. Truth.
