Kimi API Pricing (July 2026): K3 $3.00/M

Verified pricing data

Kimi API prices as of July 2026

Source: https://platform.kimi.ai/docs/pricing/chat-k3.md

As of

July 18, 2026

Update cadence

Checked daily; published only after source verification.

Source policy

Official provider pricing pages or APIs only.

Machine-readable data

/api/pricing.json

Tracked sources

https://platform.kimi.ai/docs/pricing/chat-k3.md

Estimate monthly costs with these prices

Quick pricing answers

Cheapest input

Kimi K3

$3/M input · $15/M output

Cheapest output

Kimi K3

$15/M output

Best cached-input price

Kimi K3

$0.30/M cached input

Widest context

Kimi K3

1,048,576 tokens context

These answers are generated from 1 verified Kimi models as of July 2026. Use the calculator for workload-specific totals because output tokens and cache hit rate can change the cheapest choice.

Model	Input / 1M	Cached input / 1M	Output / 1M	Context	Max output
Kimi K3	$3.00	$0.30	$15.00	1,048,576	1,048,576

Official Kimi API flat pay-as-you-go pricing: $0.30/M cache-hit input, $3/M cache-miss input, and $15/M output. K3 always reasons; reasoning tokens are billed as output. The API docs say max_completion_tokens can be set up to 1,048,576, subject to the 1M-token context limit.

Pricing questions

What is the cheapest Kimi API model in July 2026?

Kimi K3 has the lowest verified Kimi input-token price at $3/M input tokens.

Which Kimi model is cheapest for generated output?

Kimi K3 has the lowest verified Kimi output-token price at $15/M output tokens.

Where does TLDL get Kimi pricing data?

TLDL uses official Kimi pricing sources and records source URLs plus verification dates in the shared pricing dataset.

Kimi pricing changelog

2026-07-18

Model watch detected official pricing updates for kimi-k3. Review the automated draft PR before merging.

https://platform.kimi.ai/docs/pricing/chat-k3.md

2026-07-17

Added verified Kimi K3 API pricing, 1M-token context, and maximum completion-token limit from Kimi's official launch and API documentation.

https://platform.kimi.ai/docs/pricing/chat-k3

Kimi K3 uses flat pay-as-you-go pricing across its full 1-million-token context window. There is no higher long-context tier: the important billing distinction is whether an input token hits Kimi's automatic context cache.

The live table above is generated from TLDL's shared pricing dataset. The same verified row powers the LLM API pricing comparison, public pricing API, and LLM cost calculator.

Kimi K3 price at a glance

K3 always runs in thinking mode. Its reasoning tokens count as output tokens, so a short visible answer can still carry a meaningful output bill. The API currently supports only reasoning_effort="max"; Kimi says additional effort levels are coming later.

Example Kimi K3 costs

These examples use the official per-token rates and exclude taxes.

Use the cost calculator for your own monthly token volume and cache-hit assumptions.

How automatic context caching changes the bill

Kimi says context caching is automatic for normal model requests. You do not create a cache ID or choose a time-to-live value. Keep the long prefix unchanged between requests so later calls can attempt a cache hit.

That makes K3's economics especially sensitive to prompt structure:

• Put stable instructions, repository context, or a knowledge base first.
• Append changing user questions after the reusable prefix.
• Track actual cache-hit input separately from total input.
• Budget reasoning as output, not as free hidden work.

Kimi reports that its official API sees cache-hit rates above 90% in coding workloads, but your rate will depend on how often the prefix changes. Estimate conservatively until production usage confirms it.

API limits and model behavior

• Model ID: kimi-k3
• Context window: 1,000,000 tokens
• Thinking mode: always enabled
• Current reasoning effort: max only
• Maximum completion setting: up to 1,048,576 tokens, within the request's context constraints
• Vision input: supported for images and video through Kimi's documented formats
• Tool use: supports tool calls, tool_choice, and dynamic tool loading

For multi-turn chat and tool use, Kimi instructs developers to return the complete assistant message—including its reasoning history—to the next request. Dropping that history or switching models in the middle of a session can hurt K3's stability.

Is Kimi K3 cheap?

K3's $3/M uncached-input and $15/M output prices place it in the premium general-purpose API range rather than the budget-model category. The $0.30/M cached-input rate is the more distinctive number: a workload with a stable, frequently reused prefix can reduce that portion of input cost by 90%.

Compare the full market in LLM API Pricing. Quality, latency, reasoning-token volume, and cache-hit rate still need to be measured on your own workload before moving production traffic.

Related

• Kimi K3 launch: what changed
• Latest AI model releases
• LLM API pricing comparison
• LLM cost calculator