Kimi K2.7-Code API Cost: Per-Task vs DeepSeek (Tested)

Kimi K2.7-Code costs $0.74 per million input tokens and $3.50 per million output tokens when routed through OpenRouter, which makes it pricier and slower per coding task than DeepSeek V4-Flash, a gap you only justify with output quality. That is the honest verdict, and the rest of this page shows the per-task numbers we measured ourselves so you can decide whether the premium earns its keep on your workload.

If you write code with an LLM, the headline rate card hides the number that actually matters: what one real coding turn costs end to end. We pulled both. The rate card sets the ceiling; the per-task cost is what lands on your invoice.

Kimi K2.7-Code coding cost (verified 2026-06)

Kimi K2.7-Code is the coding-specialized tier of Moonshot's K2 family that pairs a 256K-token context window with a price step above the base K2 model. It is a distinct tier from the base K2 alias, and conflating the two is the most common pricing error we see.

According to Moonshot AI Platform, the K2 family is offered in international USD plans through platform.moonshot.ai, separate from the mainland RMB pricing, with the flagship K2.6 tier listed at $0.95 input and $4.00 output per million tokens (needs native re-verification). The K2.7-Code tier we benchmarked is reached through OpenRouter, and the routed rate there is $0.74 input and $3.50 output per million tokens, measured against the live OpenRouter catalog on 2026-06-26.

One clarification worth fixing up front. The OpenRouter "kimi-k2" alias resolves to the K2-0711 generation, not the flagship K2.6, so when you see base K2 priced near $0.57 input that is the older model, not the current flagship. According to Moonshot AI Platform, the long-context strength of the K2 line comes from its 262,144-token (256K) window, which is the technical reason the coding tier can hold a large repository in view (needs native re-verification). K2.7-Code sits above base K2 on price because the coding tier carries its own rate.

What one coding task actually cost (first-hand evidence)

Documentation gives you the per-million rate. It does not tell you how many tokens a real coding turn burns or how long you wait. So we called both endpoints on a coding prompt and recorded the billed cost and latency directly.

When we routed a coding prompt to K2.7-Code through OpenRouter, the served model came back as kimi-k2.7-code-20260612. That single turn read 36 input tokens, generated 201 output tokens, billed $0.00083784, and took 6.83 seconds. The same class of coding prompt sent to DeepSeek V4-Flash on its official endpoint read 27 input tokens, generated 114 output tokens, and returned in 2.35 seconds.

Two things jump out. K2.7-Code was roughly 2.9x slower on the wall clock and produced almost double the output tokens, and because its output rate is more than 12x DeepSeek's, the per-task cost gap widens fast. The practical upshot: on raw coding throughput and cost, DeepSeek V4-Flash wins this comparison cleanly. K2.7-Code only pulls ahead if its longer, more thorough output is worth the wait and the spend on your specific code.

According to DeepSeek API Docs, V4-Flash is priced at $0.14 input and $0.28 output per million tokens on the official endpoint, which is the rate we used to compute its side of the table. We should be transparent about the asymmetry: the Kimi figure is OpenRouter-routed while the DeepSeek figure is official-endpoint, so this is a tier-versus-tier coding comparison, not an identical-routing one.

Per-task cost math for a coding workload

Take a realistic agentic coding turn: about 4,000 input tokens of context plus instructions, generating 800 tokens of code and explanation. On K2.7-Code at the OpenRouter-routed rate, that is 4,000 x $0.74/1M plus 800 x $3.50/1M, which works out to roughly $0.00296 input plus $0.0028 output, about $0.0058 per turn. The same turn on DeepSeek V4-Flash at $0.14/$0.28 is roughly $0.00056 input plus $0.000224 output, about $0.00078 per turn.

That makes K2.7-Code close to 7x the per-turn cost of V4-Flash at this token mix, and our first-hand run showed it also tends to emit more output tokens, which pushes the real-world ratio higher than the rate card alone implies. At a thousand coding turns a day the gap between roughly $0.78 and roughly $5.80 daily is the part most buyers miss when they only read input prices.

When the premium is justified

Choose K2.7-Code when output quality on hard code is the bottleneck and a few seconds of extra latency per turn is acceptable, because the 256K context window lets it hold a large codebase in view and the coding tier is tuned for exactly that. Avoid it when you are running high-volume, latency-sensitive, or cost-capped coding traffic where DeepSeek V4-Flash already clears your quality bar, since on this benchmark V4-Flash was faster and far cheaper per task.

For a fuller view of how Kimi tiers compare across general and reasoning workloads, not just coding, see the cluster hub. For the DeepSeek side of the comparison in depth, the DeepSeek pricing pages cover the cache mechanics and rate card that make V4-Flash so cheap per task.

FAQ

How much does Kimi K2.7-Code cost per million tokens? Measured via OpenRouter on 2026-06-26, K2.7-Code is $0.74 per million input tokens and $3.50 per million output tokens. That is a higher tier than the base K2 alias, which routes to the older K2-0711 generation at $0.57 input and $2.30 output.

Is Kimi K2.7-Code cheaper than DeepSeek for coding? No. On our first-hand test a single coding turn on K2.7-Code billed $0.00083784 at 6.83 seconds, while DeepSeek V4-Flash returned in 2.35 seconds at $0.14/$0.28 official rates, far cheaper per task. K2.7-Code is pricier and slower here; the premium only pays off through output quality.

Which Kimi tier does the OpenRouter "kimi-k2" alias point to? The base "kimi-k2" alias resolves to the K2-0711 generation, not the flagship K2.6 and not the coding-specialized K2.7-Code. We benchmarked K2.7-Code specifically, served as kimi-k2.7-code-20260612.

Why is the Kimi price OpenRouter-routed instead of official? The native Moonshot key in our environment only reaches the legacy moonshot-v1 model, not the K2 family, so every Kimi K2 measurement here is OpenRouter-routed and disclosed as such. Official Moonshot K2 pricing is sourced from platform.moonshot.ai and still needs native re-verification.

---

This is part of the Kimi API pricing hub, and for the DeepSeek comparison baseline see the DeepSeek API pricing hub.

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

<!-- METADATA { "title": "Kimi K2.7-Code API Cost for Developers (2026)", "slug": "kimi-k2-coding-cost", "meta_description": "Kimi K2.7-Code costs $0.74/$3.50 per 1M via OpenRouter. We tested it: 201 tokens, 6.83s, $0.00083784 per turn vs DeepSeek V4-Flash. Per-task coding cost table.", "focus_keyword": "kimi k2 coding cost", "secondary_keywords": ["kimi k2.7-code pricing", "kimi k2 api cost coding", "kimi vs deepseek coding cost", "kimi k2.7 code per task cost"], "tags": ["Kimi", "Moonshot", "API Pricing", "Coding"], "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 Kimi K2.7-Code cost per million tokens?", "a": "Measured via OpenRouter on 2026-06-26, K2.7-Code is $0.74 per million input tokens and $3.50 per million output tokens. That is a higher tier than the base K2 alias, which routes to the older K2-0711 generation at $0.57 input and $2.30 output."}, {"q": "Is Kimi K2.7-Code cheaper than DeepSeek for coding?", "a": "No. On our first-hand test a single coding turn on K2.7-Code billed $0.00083784 at 6.83 seconds, while DeepSeek V4-Flash returned in 2.35 seconds at $0.14/$0.28 official rates, far cheaper per task. K2.7-Code is pricier and slower here; the premium only pays off through output quality."}, {"q": "Which Kimi tier does the OpenRouter kimi-k2 alias point to?", "a": "The base kimi-k2 alias resolves to the K2-0711 generation, not the flagship K2.6 and not the coding-specialized K2.7-Code. We benchmarked K2.7-Code specifically, served as kimi-k2.7-code-20260612."}, {"q": "Why is the Kimi price OpenRouter-routed instead of official?", "a": "The native Moonshot key in our environment only reaches the legacy moonshot-v1 model, not the K2 family, so every Kimi K2 measurement here is OpenRouter-routed and disclosed as such. Official Moonshot K2 pricing is sourced from platform.moonshot.ai and still needs native re-verification."} ], "external_links_used": [ {"url": "https://platform.moonshot.ai/", "source_name": "Moonshot AI Platform", "claim": "K2 family offered in international USD plans; flagship K2.6 listed $0.95/$4.00 (needs native re-verify)"}, {"url": "https://api-docs.deepseek.com/quick_start/pricing/", "source_name": "DeepSeek API Docs", "claim": "DeepSeek V4-Flash priced $0.14 input / $0.28 output per million tokens on official endpoint"} ], "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-k2.7-code_coding + deepseek_v4flash_coding_anchor", "measured": "K2.7-Code (served kimi-k2.7-code-20260612) via OpenRouter: 36 in / 201 out, $0.00083784, 6.83s. DeepSeek V4-Flash official anchor: 27 in / 114 out, 2.35s.", "captured": "2026-06-26", "disclosure": "Kimi K2 prices OpenRouter-routed and live-billed; native Moonshot key reaches only legacy moonshot-v1; official Moonshot K2 pricing needs native re-verification." }, "images_status": "spec-only (not generated; FAL_API_KEY unset)", "images": [ {"position": "featured", "type": "generated", "prompt": "Clean editorial bar chart comparing per-coding-task cost and latency of Kimi K2.7-Code (via OpenRouter) versus DeepSeek V4-Flash, annotated with 6.83s vs 2.35s and the measured $0.00083784 K2.7-Code turn. Indigo and teal palette, minimal background, 16:9.", "alt": "Bar chart comparing Kimi K2.7-Code and DeepSeek V4-Flash per-coding-task cost and latency, annotated with measured 6.83 seconds versus 2.35 seconds and a $0.00083784 K2.7-Code turn"} ] } -->