Kimi K2 API Pricing: Official vs OpenRouter (Verified 2026)

Base Kimi K2 costs about $0.57 per million input tokens and $2.30 per million output when routed through OpenRouter, which sits at near-parity with Moonshot's official legacy K2 listing of $0.55 and $2.20, so the third-party route is not the cheaper option here. That parity is the part most buyers get wrong, because the cheaper-third-party pattern that holds for Qwen and GLM does not hold for K2. This piece pins down exactly which K2 you are paying for, what the numbers are, and what we measured calling it ourselves.

Which Kimi K2 are you actually buying?

Kimi K2 is a long-context Chinese LLM from Moonshot AI that ships in several generations under one confusing name, so the first job is matching the price to the tier. The "kimi-k2" alias most third-party routers expose is the K2-0711 generation, the original base model. It is not the current flagship, which is K2.6, and conflating the two is the single most common pricing mistake we see in buyer questions.

The practical upshot: when you read "$0.57 in / $2.30 out" you are looking at base K2-0711 routed through OpenRouter, and when you read "$0.95 in / $4.00 out" you are looking at the flagship K2.6, a higher tier with a longer context window. Same family, different bill.

According to Moonshot AI Platform, the K2 line is offered internationally in USD plans through platform.moonshot.ai, distinct from the mainland RMB pricing, so US and international buyers can reach it directly without a reseller.

Kimi K2 pricing table (verified 2026-06)

The table below separates the OpenRouter-routed figure we can bill against today from the official Moonshot figures, which we sourced rather than measured natively and which carry a pending re-verification flag.

Tier	Source	Input ($/1M)	Output ($/1M)	Notes
K2 base (K2-0711)	measured via OpenRouter	$0.57	$2.30	the "kimi-k2" alias
K2 (0711, legacy)	official Moonshot	$0.55	$2.20	near-parity with the route above
K2.5	official Moonshot	$0.60	$3.00	mid generation
K2.6 (current flagship)	official Moonshot	$0.95	$4.00	cached input $0.19, 256K context

According to Moonshot AI Platform, the flagship K2.6 carries a 262,144-token (256K) context window with cached input billed at $0.19 per million, which is the long-context strength the K2 family is known for. The base K2 alias does not share that flagship rate card, so size your budget against the specific tier your code requests, not the family headline.

For readers comparing across vendors, it helps to anchor against a model whose official rates we can cite directly. According to DeepSeek API Docs, DeepSeek V4-Flash bills input at $0.14 and output at $0.28 per million, which is several times below base K2 either way you route it. The trade is context and capability, not price, so the comparison only makes sense once you know what each model is for.

Is OpenRouter cheaper for K2? No, and that is worth saying plainly

Here is the honest finding, because forcing the usual narrative would mislead you. For base K2 the OpenRouter-routed price of $0.57 in and $2.30 out is a slight premium over the official legacy K2 listing of $0.55 in and $2.20 out, not a discount. The third-party route buys you convenience, a unified billing account, and easy model switching. It does not buy you a lower per-token rate on this model.

That is the opposite of the pattern buyers learn from Qwen and GLM, where third-party routing often undercuts the official endpoint. K2 breaks the rule. If your only goal is the lowest base-K2 token price and you can reach platform.moonshot.ai, the official route edges it. If you value one account across many models, the few cents of premium is the toll for that convenience.

We called K2 ourselves (first-hand evidence)

Documentation gives you a rate card. We wanted real token counts, real latency, and a real billed cost, so we called base K2 directly. These numbers were measured via OpenRouter, since the native Moonshot key in our environment only reaches the legacy moonshot-v1 endpoint and not K2; the official Moonshot K2 figures above remain sourced and pending native re-verification.

On a short general-purpose prompt, base K2 (the moonshotai/kimi-k2 alias) read 36 input tokens, produced 60 output tokens, billed $0.00015852, and returned in 7.03 seconds. On a small coding prompt the same model read 35 in, wrote 41 out, billed $0.00011425, and came back faster at 4.55 seconds.

Test	Input tok	Output tok	Billed cost	Latency
K2 general	36	60	$0.00015852	7.03s
K2 coding	35	41	$0.00011425	4.55s

What stands out is how cheap a single base-K2 call is in absolute terms: a fraction of a cent for a complete response. The latency is the variable to watch, not the per-call cost. Seven seconds on a trivial prompt is unremarkable for a quality-tier model but matters if you are chaining dozens of calls in an agent loop. We re-ran both to be sure the figures held, and they did.

One caution that comes straight from our runs: the reasoning tier is a different animal. The same two-sentence prompt sent to K2-thinking produced 553 output tokens at 18.5 seconds and roughly nine times the cost of base K2. If you do not need step-by-step reasoning, base K2 is the calmer, cheaper default.

Where K2 fits against the cheaper alternatives

Choose base K2 when you want Moonshot's long-context lineage at a mid-tier price and you value capability over the rock-bottom rate. Avoid paying the K2.6 flagship rate when base K2-0711 already covers the job, since that is a real step up in cost for the longer context you may not use. And if your workload is price-sensitive and context-light, a model like DeepSeek V4-Flash at $0.14 / $0.28 will run several times cheaper. For that side-by-side rate math, see our DeepSeek API pricing hub.

FAQ

How much does Kimi K2 cost per million tokens? Base K2 (the K2-0711 generation) is about $0.57 input and $2.30 output per million when routed through OpenRouter, and roughly $0.55 / $2.20 on Moonshot's official legacy K2 listing. The flagship K2.6 is a higher tier at $0.95 / $4.00 official.

Is the base "kimi-k2" alias the same as the flagship K2.6? No. The kimi-k2 alias most routers expose is the base K2-0711 generation. K2.6 is the current flagship, priced higher at $0.95 / $4.00 with a 256K context window. Do not budget one against the other.

Is OpenRouter cheaper than Moonshot for K2? No. For base K2 the OpenRouter route ($0.57 / $2.30) is a slight premium over the official legacy K2 listing ($0.55 / $2.20). You pay a few cents extra for unified billing and easy model switching, not less per token.

What did a real K2 call cost when you tested it? Measured via OpenRouter, a general prompt on base K2 read 36 input and 60 output tokens, billed $0.00015852, and took 7.03 seconds. A coding prompt billed $0.00011425 at 4.55 seconds.

Is Kimi K2 available to US and international buyers? Yes. According to Moonshot's platform, K2 is offered internationally in USD plans through platform.moonshot.ai, separate from the mainland RMB pricing.

This is part of the Kimi API pricing hub, which compares every K2 tier alongside other Chinese-model options.

Author: Kevin Fan, Customer Success Manager at China LLM Directory, specializing in Chinese LLM ecosystem pricing. Last verified: 2026-06-26.