Kimi 256K Long-Context Cost: What One Full Window Costs

Filling Kimi's full 256K context window once costs about $0.19 in input tokens alone on the K2.6 flagship, before you pay for a single output token, because at $0.95 per million input tokens a 200K-token prompt is 0.2 million tokens times that rate. That number is the whole reason long-context pricing deserves its own look: the window is Kimi's signature feature, and it is also where the bill quietly grows.

What a Kimi long-context call actually costs (verified 2026-06)

Kimi's defining trait is its context window. Kimi K2.6 is a long-context large language model that exposes a 262,144-token (256K) window, the same window carried by K2.7-Code, and that headroom is what people reach for when they paste an entire codebase or a stack of contracts into one prompt. The catch is that you pay for every token you put in.

According to Kimi/Moonshot pricing on the international platform, K2.6 lists at $0.95 per million input tokens and $4.00 per million output tokens, with cached input at $0.19 per million and the 262,144-token context. These are official figures sourced from Moonshot and carry a needs_native_reverify flag, because our native Moonshot key in this environment only reaches the legacy moonshot-v1 endpoint, not the K2 generation.

Here is the arithmetic that matters. Fill 200K tokens of input, which is most of the window, and you spend 0.2 million times $0.95, or $0.19, on input before the model writes anything back. Push to the full 256K and input alone is roughly $0.25. The practical upshot is that with a 256K model your prompt size, not your output length, often drives the bill.

The cached row is the one worth staring at. According to Kimi/Moonshot pricing, cached input on K2.6 bills at $0.19 per million, a 5x discount on the uncached rate. If you send the same 200K of reference context across many calls, the second call onward pays $0.038 to reload that context instead of $0.19, which is the difference between context being a per-call tax and a one-time setup cost.

First-hand: what base K2 billed us, measured via OpenRouter

We wanted a hard per-token anchor that was not lifted from a pricing page, so we called Kimi directly. One caveat up front, disclosed plainly: the native Moonshot key in this environment only reaches legacy moonshot-v1, so to touch the K2 generation we routed through OpenRouter. Every Kimi number in this section is the OpenRouter-routed, live-billed value.

The model we tested is the OpenRouter kimi-k2 alias, which is the K2-0711 generation, not the K2.6 flagship. That distinction matters for budgeting and we keep it explicit throughout.

On a short general-purpose prompt, base K2 read 36 input tokens, wrote 60 output tokens, billed $0.00015852, and returned in 7.03 seconds. That is a tiny call by design, the point being the per-token rate it implies, not the absolute figure. The base K2 alias prices at $0.57 input and $2.30 output per million via OpenRouter, which sits at near-parity with the official legacy K2 rate of $0.55 / $2.20. We will say the honest thing here that some roundup pages avoid: for base K2, routing through OpenRouter is not cheaper than going official, it is a slight premium. Treat the OpenRouter figure as a convenience price, not a discount.

The reason this anchor is useful for long context is scaling. The base K2 alias is the cheap tier; the 256K window lives on the pricier K2.6 flagship at $0.95 / $4.00. So a long-context call on the model that actually offers 256K costs meaningfully more per token than the base-K2 measurement above, and you should budget against K2.6's rate, not the base alias, whenever the window is the reason you chose Kimi.

When the 256K window is worth the bill vs chunking on a cheaper model

The real decision is rarely "Kimi yes or no." It is "one big 256K call, or several small calls on something cheaper." The math turns on how much of that giant context you reuse.

If you load 200K of context once and ask a single question, you pay roughly $0.19 of input on K2.6 and move on. That is cheap for a one-shot deep read. If you load the same 200K and ask forty questions across a session without caching, you pay that $0.19 forty times, about $7.60 of input, and chunking starts to look smart. With K2.6's cached input at $0.19 per million, the reused context drops to $0.038 per follow-up call, which is what keeps a long-context chat affordable.

The alternative is chunking on a cheaper model. According to DeepSeek API pricing, DeepSeek V4-Flash bills input at $0.14 per million, roughly 7x below K2.6's $0.95. If your task splits cleanly into independent chunks that each fit a smaller window, running them through a cheaper model and stitching the answers can undercut a single 256K Kimi call by a wide margin. We have the DeepSeek side measured too: our V4-Flash coding anchor, called against the official api.deepseek.com endpoint, read 27 input tokens and wrote 114 output in 2.35 seconds. Note the asymmetry honestly, our Kimi numbers are OpenRouter-routed while the DeepSeek anchor is official-endpoint, so treat the comparison as directional on rate, not a controlled head-to-head.

So the verdict splits cleanly. Choose the 256K window when the task genuinely needs whole-document reasoning that chunking would break, cross-references across a long contract, a full codebase, a book-length transcript, and especially when you can cache the shared context across many calls. Avoid paying for 256K when your work decomposes into independent pieces, because there a cheaper model run several times will almost always cost less than one window-filling Kimi call.

According to Kimi/Moonshot pricing, the 262,144-token window is available on K2.6 and K2.7-Code, so the long-context capability is not limited to a single SKU; the cost model above applies wherever you use the full window.

For the full rate card across every Kimi tier, the Kimi API pricing hub collects them in one place, and the DeepSeek API pricing hub covers the cheaper-chunking comparison in depth.

FAQ

How much does it cost to fill Kimi's 256K context window? Filling 200K of the 256K window costs about $0.19 in input tokens on the K2.6 flagship at $0.95 per million, before any output. The full 256K is roughly $0.25 of input. Output is billed separately at $4.00 per million.

Does Kimi charge less for cached long context? Yes. According to Moonshot's official pricing, K2.6 cached input bills at $0.19 per million versus $0.95 uncached, a 5x discount. Reusing a large shared context across many calls is what keeps a long-context Kimi workload affordable.

Which Kimi model has the 256K context window? The 262,144-token (256K) window is on the K2.6 flagship and on K2.7-Code, according to Moonshot pricing. The cheaper base K2 alias is a separate, lower tier, so budget long-context work against K2.6's $0.95 / $4.00 rate, not the base alias.

Is it cheaper to use Kimi's big window or chunk on another model? It depends on reuse. One-shot whole-document reads favor the 256K window. Many independent queries favor chunking on a cheaper model such as DeepSeek V4-Flash at $0.14 input per million, unless you can cache Kimi's shared context to cut reload cost to $0.038 per call.

Are these Kimi prices measured or quoted? The base-K2 per-token anchor ($0.00015852 on a 36-in/60-out call) was measured first-hand via OpenRouter on the K2-0711 alias. The K2.6 flagship rates ($0.95 / $4.00, cached $0.19) are official Moonshot figures pending native re-verification.

---

This article is part of the Kimi API pricing hub, where every Kimi tier and its long-context cost model are collected.

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

<!-- METADATA { "title": "Kimi 256K Long-Context API Cost Guide (2026)", "slug": "kimi-long-context-cost", "meta_description": "Filling Kimi's 256K window costs ~$0.19 input alone on K2.6 at $0.95/M. When that long-context bill beats chunking on a cheaper model, with first-hand K2 data.", "focus_keyword": "kimi long context cost", "secondary_keywords": ["kimi 256k context price", "kimi k2.6 input cost", "kimi long context api pricing", "kimi context window cost"], "tags": ["Kimi", "API Pricing", "Long Context"], "category": "Pricing", "cluster_id": "kimi-api-pricing", "cluster_role": "micro", "hub_slug": "kimi-api-pricing", "evidence_file": "clients/china-llm-aggregator/articles/kimi-api-pricing-evidence.json", "needs_native_reverify": true, "verified_until": "2026-09-24", "author_name": "Kevin Fan", "author_title": "Customer Success Manager", "author_linkedin": "", "author_expertise": ["Chinese LLM ecosystem", "AI infrastructure pricing", "model benchmarking", "cross-border AI compliance"], "faq_pairs": [ {"q": "How much does it cost to fill Kimi's 256K context window?", "a": "Filling 200K of the 256K window costs about $0.19 in input tokens on the K2.6 flagship at $0.95 per million, before any output. The full 256K is roughly $0.25 of input. Output is billed separately at $4.00 per million."}, {"q": "Does Kimi charge less for cached long context?", "a": "Yes. According to Moonshot's official pricing, K2.6 cached input bills at $0.19 per million versus $0.95 uncached, a 5x discount. Reusing a large shared context across many calls is what keeps a long-context Kimi workload affordable."}, {"q": "Which Kimi model has the 256K context window?", "a": "The 262,144-token (256K) window is on the K2.6 flagship and on K2.7-Code, according to Moonshot pricing. The cheaper base K2 alias is a separate, lower tier, so budget long-context work against K2.6's $0.95 / $4.00 rate, not the base alias."}, {"q": "Is it cheaper to use Kimi's big window or chunk on another model?", "a": "It depends on reuse. One-shot whole-document reads favor the 256K window. Many independent queries favor chunking on a cheaper model such as DeepSeek V4-Flash at $0.14 input per million, unless you can cache Kimi's shared context to cut reload cost to $0.038 per call."}, {"q": "Are these Kimi prices measured or quoted?", "a": "The base-K2 per-token anchor ($0.00015852 on a 36-in/60-out call) was measured first-hand via OpenRouter on the K2-0711 alias. The K2.6 flagship rates ($0.95 / $4.00, cached $0.19) are official Moonshot figures pending native re-verification."} ], "external_links_used": [ {"url": "https://platform.moonshot.ai/", "source_name": "Kimi/Moonshot pricing", "claim": "K2.6 $0.95 in / $4.00 out, cached input $0.19, 262144-token (256K) context on K2.6 and K2.7-Code"}, {"url": "https://api-docs.deepseek.com/quick_start/pricing/", "source_name": "DeepSeek API Docs – Pricing", "claim": "DeepSeek V4-Flash input $0.14/M, used as cheaper-chunking comparison anchor"} ], "internal_links_used": [ {"url": "/blog/kimi-api-pricing/", "anchor_text": "Kimi API pricing hub", "type": "hub"}, {"url": "/blog/deepseek-api-pricing/", "anchor_text": "DeepSeek API pricing hub", "type": "cross-cluster"} ], "first_hand_evidence": { "source": "kimi-api-pricing-evidence.json run label kimi-k2_general (+ deepseek_v4flash_coding_anchor)", "measured": "base K2 (OpenRouter kimi-k2 alias = K2-0711): 36 in / 60 out, billed $0.00015852, 7.03s, measured via OpenRouter; DeepSeek V4-Flash anchor: 27 in / 114 out, 2.35s, official api.deepseek.com", "routing_disclosure": "Kimi K2 routed via OpenRouter; native Moonshot key only reaches legacy moonshot-v1. DeepSeek anchor via official endpoint. needs_native_reverify.", "captured": "2026-06-26" }, "images_status": "spec-only (not generated; FAL_API_KEY unset)", "images": [ {"position": "featured", "type": "generated", "prompt": "Clean editorial diagram of a 256K token context window filling up, annotated with the cost meter rising to $0.19 of input on Kimi K2.6 before any output token, with a cached-context path showing the 5x cheaper $0.038 reload. Indigo and amber palette. Minimal background. 16:9.", "alt": "Diagram of Kimi's 256K context window filling with tokens, annotated showing $0.19 input cost on K2.6 before output and a 5x cheaper cached reload path"} ] } -->