Quick answer. Kimi K2.6 costs $0.95 per million input tokens on a cache miss, $0.16 on a cache hit, and $4.00 per million output tokens from Moonshot's own API, with a 262,144-token context window. Every figure here was read from a vendor page on 2026-10-01 and is dated for that reason — this model's rates have moved more than once.
The number most guides miss is the second one. On Kimi's own API the cache-hit input price is six times cheaper than the cache-miss price, which makes "how much does Kimi K2.6 cost" a question about your prompt structure more than about the rate card.
Quick comparison
| Model | Input (cache hit) | Input (cache miss) | Output | Context |
|---|---|---|---|---|
| kimi-k2.6 | $0.16 | $0.95 | $4.00 | 262,144 |
| kimi-k2.7-code | $0.19 | $0.95 | $4.00 | 262,144 |
| kimi-k2.7-code-highspeed | $0.38 | $1.90 | $8.00 | 262,144 |
| kimi-k3 | $0.30 | $3.00 | $15.00 | 1,048,576 |
All prices per million tokens, read from platform.kimi.ai pricing on 2026-10-01.
Kimi K2.6 pricing: the official API rates
Moonshot publishes three numbers per model rather than the usual two, and the split is the whole story:
- Input, cache hit — $0.16 / MTok. What you pay for tokens the service has already processed and retained.
- Input, cache miss — $0.95 / MTok. What you pay for tokens it is seeing fresh.
- Output — $4.00 / MTok. What the model generates.
K3, the larger sibling, also publishes cache-write prices: $3.00 per million for a 5-minute TTL and $6.00 for one hour, with cached input at $0.30 and fresh input at $3.00. K2.6's listing does not carry a separate cache-write line.
The cache-hit price is the number that matters
A 6× gap between cache-hit and cache-miss input is larger than the gap between most competing models. It means two applications paying the same published rate can have bills that differ by several times.
Work out which side of that line you sit on before comparing Kimi to anything else:
You will mostly hit the cache if you re-send a long stable prefix — a system prompt, a document, a codebase, a conversation that grows by one turn at a time. Agent loops and chat sessions are the natural case.
You will mostly miss if every request carries different content — batch classification over unrelated documents, one-shot extraction, anything fan-out.
At $0.16 the input side almost stops mattering and output dominates. At $0.95 input and output are closer to comparable, and a verbose model costs you twice.
How much does Kimi K2.6 cost in practice
Three worked examples, using the published rates and nothing else:
| Workload | Input | Output | Cost |
|---|---|---|---|
| One long chat turn, 50k cached context + 2k new, 1k reply | 50k hit + 2k miss | 1k | $0.014 |
| Document extraction, 200k fresh input, 5k out | 200k miss | 5k | $0.210 |
| Agent run, 20 steps × (80k cached + 3k new + 2k out) | 1.6M hit + 60k miss | 40k | $0.473 |
The agent row is the one worth staring at. Twenty steps across 1.66 million input tokens lands under fifty cents because almost all of it is cached. The same run with no cache reuse would cost $1.74 — 3.7× more for identical work.
It is cheaper through OpenRouter than from Moonshot
This is the finding most comparisons do not have, and it is checkable in a browser.
Read from OpenRouter's public models API on 2026-10-01:
| Route | Input | Output |
|---|---|---|
| Moonshot direct (cache miss) | $0.95 | $4.00 |
OpenRouter moonshotai/kimi-k2.6 | $0.434 | $1.828 |
Output through OpenRouter is 54% below the vendor's own published rate, and input is less than half. Context is the same 262,144 either way.
The reason is routing: OpenRouter sends requests to providers that host the model rather than to Moonshot, and those providers compete on price. Two caveats before you move:
- OpenRouter's rate card has no cache-hit tier for this model. If your workload is cache-heavy, Moonshot's $0.16 beats OpenRouter's $0.434 — the direct route wins by 2.7× on input. The comparison flips entirely depending on your reuse pattern.
- Routed prices change without an announcement, because they track whichever provider is serving you. A vendor page is a commitment; a routing marketplace is a quote.
So the honest rule: cache-heavy work goes direct, fresh-input work goes through the marketplace.
K2.6 is not the newest model any more
Anyone arriving at this page from a six-month-old recommendation should know the lineup moved:
| Model | Input (miss) | Output | Context |
|---|---|---|---|
| kimi-k2.6 | $0.95 | $4.00 | 262,144 |
| kimi-k2.7-code | $0.95 | $4.00 | 262,144 |
| kimi-k3 | $3.00 | $15.00 | 1,048,576 |
K2.7-code costs the same as K2.6 on a cache miss and differs only in the cache-hit tier ($0.19 against $0.16). If you are starting a coding workload today there is no price reason to pick the older one.
K3 costs 3.2× the input and 3.75× the output and gives you four times the context. That is a real trade, not an upgrade — a million-token window is worth paying for when you genuinely load a million tokens, and is dead weight when you do not.
Is Kimi K2.6 worth it?
At $0.95 / $4.00 it sits below the frontier Western models on output price while offering a 262k window, so for high-volume generation the arithmetic is straightforward. The cases where it is not the right call:
- You need a million-token context. K2.6 stops at 262,144. That is K3's job, at 3.75× the output price.
- Your workload is cache-heavy and you want the marketplace price. You cannot have both: OpenRouter is cheaper on fresh input, Moonshot is far cheaper on cached input. Pick by pattern, not by headline.
- You need a published commitment rather than a routed quote. Marketplace pricing moves without notice.
- You are starting new coding work. K2.7-code is the same cache-miss price with a newer target.
How it compares with Claude
For a cross-vendor read at the same moment: Claude Sonnet 5 is $2.00 input and $10.00 output per million tokens, and Opus 5.5 is $4.00 / $20.00 — figures dated 2026-09-26 in our Claude Code pricing breakdown.
On output, K2.6 at $4.00 undercuts Sonnet 5 by 60% and Opus 5.5 by 80%. On cache-miss input it is roughly half Sonnet 5's rate. The window is the trade: 262,144 tokens against Claude's larger context on the top tiers.
If you are pricing image or video generation rather than text, the same exercise for xAI is in our Grok Imagine pricing page — that one bills per image and per second, not per token.
What we could not verify
Kimi's consumer subscription tiers. kimi.com renders its plan names and prices client-side, and they are not present in the page source we could read on 2026-10-01. Other articles quote figures for a Kimi membership; treat those as unverified until you see them on Moonshot's own page.
We would rather leave a gap than fill it with a number copied from someone else's post. Everything above is an API rate, read from the vendor or from OpenRouter's public API, and dated.
FAQ
How much does Kimi K2.6 cost?
$0.95 per million input tokens on a cache miss, $0.16 per million on a cache hit, and $4.00 per million output tokens, with a 262,144-token context window — read from platform.kimi.ai on 2026-10-01. Through OpenRouter the same model is $0.434 input and $1.828 output, but with no cache-hit tier.
What is Kimi K2.6's context window?
262,144 tokens, the same as kimi-k2.7-code. The larger kimi-k3 offers 1,048,576 tokens at $3.00 input and $15.00 output per million.
Why are there two input prices for Kimi K2.6?
Moonshot bills cached and uncached input separately. Tokens the service has already processed cost $0.16 per million; fresh tokens cost $0.95. The 6× gap means prompt structure affects your bill more than the headline rate does.
Is Kimi K2.6 cheaper than Kimi K3?
Substantially. K2.6 is $0.95 input and $4.00 output per million against K3's $3.00 and $15.00 — 3.2× and 3.75× respectively. K3 buys you a 1,048,576-token context instead of 262,144, which is worth paying for only if you actually use it.
Is Kimi K2.6 cheaper on OpenRouter?
For fresh input, yes: $0.434 against $0.95, and $1.828 output against $4.00 — 54% less on output. For cached input, no: Moonshot's $0.16 beats OpenRouter's $0.434 by 2.7×, because OpenRouter publishes no cache-hit tier for this model. Which route is cheaper depends entirely on how much of your prompt repeats.
Is Kimi K2.6 worth it?
For high-volume generation on a 262k window, the output price is the argument. It is the wrong choice if you need a million-token context (use K3), if you are starting new coding work (K2.7-code is the same cache-miss price), or if you need a price commitment rather than a marketplace quote.
How does Kimi K2.6 compare with kimi-k2.7-code?
Identical on cache-miss input ($0.95) and output ($4.00), and nearly identical on cache hits ($0.16 against $0.19). Both carry a 262,144-token context. There is no meaningful price gap between them.
How were these prices verified?
Moonshot's own pricing page at platform.kimi.ai/docs/pricing/chat and OpenRouter's public models API were both read on 2026-10-01, and the figures are quoted as published rather than converted or estimated. Kimi's consumer subscription prices render client-side and could not be read, so none are quoted here. Rates in this category change quickly — confirm on the vendor's page before committing.