Kimi API Pricing: K3, K2.7 Code and K2.6 Costs

Kimi API uses pay-as-you-go billing. There is no Open Platform subscription plan: you add credit, create an API key and pay for the tokens and paid tools your requests use.

This page covers the three models prominently listed by the official Kimi API Platform on August 4, 2026. It also explains cached input, web-search charges, Batch API pricing and the difference between API credit and Kimi Membership.

Quick answer: Kimi K3 costs $0.30 per million cached input tokens, $3.00 per million uncached input tokens and $15.00 per million output tokens. Kimi K2.7 Code costs $0.19 / $0.95 / $4.00 for the same categories. Kimi K2.6 costs $0.16 / $0.95 / $4.00. Taxes are excluded.

Current Kimi API price table

Model	Cache-hit input	Cache-miss input	Output	Context window
`kimi-k3`	$0.30 / 1M tokens	$3.00 / 1M tokens	$15.00 / 1M tokens	1M tokens
`kimi-k2.7-code`	$0.19 / 1M tokens	$0.95 / 1M tokens	$4.00 / 1M tokens	256K tokens
`kimi-k2.7-code-highspeed`	$0.38 / 1M tokens	$1.90 / 1M tokens	$8.00 / 1M tokens	256K tokens
`kimi-k2.6`	$0.16 / 1M tokens	$0.95 / 1M tokens	$4.00 / 1M tokens	256K tokens

Here, 1M means 1,000,000 tokens. Kimi says listed prices exclude applicable taxes. The final charge can combine all three token categories in one request.

The high-speed K2.7 Code variant is the same coding-focused model served at a higher advertised output speed and double the standard K2.7 Code token rates. Confirm its live rate and availability before production use because model lists and price displays can change.

How the Kimi API bill is calculated

For a standard chat-completion request:

cost =
  (cache_hit_input_tokens / 1,000,000 × cache_hit_rate)
+ (cache_miss_input_tokens / 1,000,000 × input_rate)
+ (output_tokens / 1,000,000 × output_rate)
+ paid_tool_fees

Document text passed to a model counts as input. The official documentation says file extraction and storage interfaces are temporarily free, but extracted content becomes billable when included in a model request.

Worked price examples

These calculations illustrate the published rates; they are not measurements from a paid live test.

Example 1: a small K2.6 request without caching

Assume 20,000 uncached input tokens and 5,000 output tokens:

Input:  20,000 / 1,000,000 × $0.95 = $0.019
Output:  5,000 / 1,000,000 × $4.00 = $0.020
Total:                                      $0.039

Taxes and any paid tools are not included.

Example 2: the same K2.6 input hits the cache

Assume all 20,000 input tokens qualify for the cache-hit rate:

Input:  20,000 / 1,000,000 × $0.16 = $0.0032
Output:  5,000 / 1,000,000 × $4.00 = $0.0200
Total:                                       $0.0232

Caching lowers the input portion, not the output portion.

Example 3: a K3 long-context request

Assume 500,000 uncached input tokens and 50,000 output tokens:

Input:  500,000 / 1,000,000 × $3.00  = $1.50
Output:  50,000 / 1,000,000 × $15.00 = $0.75
Total:                                        $2.25

A 1M-token context window is a capacity limit, not a flat fee. You pay for the tokens actually processed.

Calculate a custom Kimi API estimate

The examples above explain the billing formula. To change the model, token mix, request count, cache assumptions, Batch setting or web-search usage, use the Kimi API Cost Calculator.

The calculator runs in the browser and does not require an API key. Its output is a planning estimate based on the documented rates checked for the page; actual usage, cache recognition, taxes, tool behavior and future price changes can alter the final charge.

Context caching

Automatic context caching discounts input that the platform recognizes as reusable context. It can help when requests repeat a long system prompt, codebase prefix or reference document.

Important limitations:

A cache hit is not guaranteed for every repeated request.
Only the hit portion receives the lower input rate.
Generated output is still billed at the normal output rate.
Reordering or changing context can affect reuse.

Use the token-usage fields returned by the API and your console’s usage records rather than forecasting every request as a full cache hit.

Web-search pricing: an official source conflict

The current Open Platform WebSearch guide says a triggered $web_search call costs $0.005, plus the normal model tokens. Search-result content can also add input tokens to the continuation request.

Kimi’s current Help Center API pricing article says $0.004 per invocation. These two first-party pages conflict, and we have not completed a paid billing observation that resolves the difference. We therefore do not select either figure as a single current rate. The calculator shows a $0.004-$0.005 per-call scenario range whenever web search is included. Check the live console and an actual usage record before budgeting a large number of searches.

No tool-call fee is charged when the model response stops without actually triggering $web_search, according to the platform page.

Batch API pricing

Kimi states that Batch API inference costs 60% of the standard model price for supported models. The current page lists kimi-k2.7-code, kimi-k2.6 and kimi-k2.5 as supported. It does not list K3.

Batch work is intended for bulk jobs that do not need real-time completion. Jobs must finish within their specified completion window, and the Batch API has its own operational constraints.

Recharge requirements and rate limits

The Open Platform currently says an account must recharge at least $1 before it can start using the API. Rate limits depend on cumulative recharge and may also be adjusted under platform load or risk controls.

Rate-limit measures can include:

concurrent requests;
requests per minute (RPM);
tokens per minute (TPM); and
tokens per day (TPD).

Credit balance and rate-limit capacity are different. A positive balance does not guarantee unlimited concurrency.

API vs Membership vs Kimi Code

If you need…	Use…
Kimi’s hosted personal chat and Agent workspace	Kimi Membership
A ready-made coding agent in terminal or IDE	Kimi Code membership benefit
Programmatic model calls inside your own application	Kimi API Open Platform
Managed team seats and an isolated organization workspace	Kimi Business

Kimi explicitly states that Open Platform billing is separate from Membership and Kimi Code. Do not buy a membership expecting a transferable API balance.

Compare all product types in our Kimi AI pricing hub or start integrating with the Kimi API guide.

How to control Kimi API spending

Choose the least expensive model that passes your own quality test. Do not select only by benchmark rank.
Estimate tokens before sending unusually large requests. Kimi documents a token-estimation endpoint.
Set output limits. Long completions can dominate cost, especially with K3.
Reuse stable prefixes. Structure repeated context to improve the chance of cache hits.
Log usage fields. Record prompt, cached and completion token totals without logging secrets or sensitive prompts.
Use Batch for suitable offline work. Its discount is useful only when latency and model support fit the task.
Treat tool calls as separate billable events. A web search adds both a tool fee and potentially more token input.
Create separate keys for projects. This improves attribution and makes revocation safer.

Frequently asked questions

Is there a monthly Kimi API plan?

No. Kimi’s Open Platform documentation describes usage-based, pay-as-you-go billing, with enterprise arrangements available through sales.

Does Kimi give free API credit?

Promotions and vouchers can change. The current rate-limit page says at least a $1 recharge is required to start using the API and describes a voucher at a cumulative $5 recharge. Verify any promotion in your own console before relying on it.

Which Kimi API model is cheapest?

Among the three models compared here, K2.6 has the lowest cache-hit input rate. K2.6 and standard K2.7 Code share the same listed cache-miss input and output rates. Total cost still depends on tokens, cache behavior and tools.

Does Kimi charge for files?

Kimi says file extraction and storage interfaces are temporarily free, but document content is charged as input when you send it to a model. “Temporarily” means this policy can change.

Verification and limitations

KI AI Team checked the official Kimi API homepage and pricing documentation on August 5, 2026. We did not execute a paid request for this update. The examples are arithmetic using published rates, not observed invoices. The unresolved web-search fee is shown as a range rather than a single rate. Send a correction through Sources & Corrections.

Kimi API Pricing Explained