Kimi K3.1 Pro API Pricing: China Unlock, Membership, vs Claude & DeepSeek

I keep seeing people paste one screenshot of Kimi K3.1 Pro API pricing and treat it like a universal sticker. It is not. China is in yuan on platform.moonshot.cn. International is in dollars on platform.moonshot.ai. Membership is a shared credit pool on kimi.com / kimi.ai. Claude and DeepSeek are third bills with their own docs. Mix any two and your spreadsheet lies.

I re-checked the official pages on 2026-10-24 (Asia/Shanghai). Numbers below are from those pages, not from a random X crop. Where Moonshot does not publish an absolute monthly message or token quota, I mark it missing instead of inventing one. Qwen 4 still has no official list price I can cite, so this piece does not invent one.

Kimi K3.1 Pro API pricing in one screen

Per million tokens (MTok), checked 2026-10-24:

Model	Track	Input	Output	Cache hit	5m cache write	1h cache write
k3.1-pro	China ¥	13	68	1.30	13	26
k3.1-pro-highspeed	China ¥	26	136	1.30	26	52
k3.1-pro	Intl $	2.00	10.00	0.20	2.00	4.00
k3.1-pro-highspeed	Intl $	4.00	20.00	0.20	4.00	8.00

Sources: Moonshot China pricing and Moonshot international pricing. Highspeed roughly doubles input and output; cache-hit stays on the same cheap row as the normal Pro. kimi-k3 sits nearby at ¥20/¥100 or $3/$15 if you need a sibling row, but this article is about K3.1 Pro.

I am not converting ¥ to $ with a made-up FX rate and calling that “official.” Two currencies, two invoices.

China: ¥10 unlock, and the ¥15 voucher does not help

This is the part that burns new China accounts. The official K3.1 Pro quickstart says you unlock the model after a top-up, minimum ¥10. The ¥15 signup voucher after registration and verification cannot be used on Kimi K3.1 Pro. Same FAQ on that page: no, the gift voucher does not buy you a free try of this model.

Rate limits still climb with cumulative paid recharge. Recharge and rate limits is clear that vouchers do not count toward that total. Tier0 starts tiny; Tier1 opens at ¥50 cumulative, then ¥100 / ¥500 / ¥5,000 / ¥20,000. If you only burn vouchers, you stay soft-capped while wondering why RPM feels stuck.

International quickstart is shorter: unlock after a successful top-up, minimum $1. I did not find the same “gift voucher banned on K3.1 Pro” sentence on the EN page, so I am not inventing a mirror rule.

Membership is not your API meter

People ask “which Plus plan includes how many million K3.1 Pro tokens?” Moonshot’s help center does not publish absolute monthly message or token numbers for the new tiers. It publishes relative Agent multiples and feature gates. So absolute quotas here: missing. Do not let a blog invent “Plus = 50M tokens.”

China monthly (help center basis dated 2026-10-19; buy page wins if they diverge), from kimi.com membership pricing:

Plan	Monthly	Annual total	Effective / month on annual
Go	¥49	¥468	¥39
Plus	¥99	¥948	¥79
Pro	¥199	¥1,908	¥159
Max	¥699	¥6,708	¥559
Ultra	¥1,399	¥13,428	¥1,119

International (kimi.ai membership pricing): Plus $19 / Pro $39 / Max $99 / Ultra $199 monthly. No Go tier. Annual discounts exist; again, purchase page is authoritative.

For Kimi Code model access, the Code docs are the ones that matter: ordinary K3.1 Pro / K3 calls need Plus+ on the new plans; 1M context and Highspeed (including k3.1-pro-highspeed) need Pro+. Code shares the same monthly pool as the membership web/CLI/VS Code/API-key path. New plans dropped the weekly limit and kept a rolling 5-hour usage cap — that is not “only five hours a week.” Exact token numbers inside that 5-hour window: also missing from the help center (console shows your own bar). Extra Usage top-ups exist after the pool runs out; exact Extra Usage per-MTok table: missing (official wording is “approx. open-platform API”).

Membership bill ≠ open-platform K3.1 Pro bill. One is a credit pool with gates. The other is the ¥/$ table above.

Kimi K3.1 Pro API pricing vs Claude and DeepSeek

Same date check. I put international Kimi next to Claude Global and DeepSeek’s English Peak cache-miss row so the units match. China DeepSeek yuan and Claude US-only ×1.1 are footnotes, not the same column.

Claude (platform.claude.com pricing, USD / MTok, Global):

Model	Input	Output	Notes
Claude Opus 5.5	$4	$20	Fast mode $8/$40; Batch half; cache write/read listed separately
Claude Sonnet 5.5	$2	$10	Cache hit 0.05× input
Claude Haiku 5.5	$0.10 / $0.50	$0.50 / $2.50	Split by prompt ≤100K vs >100K

US-only (inference_geo: "us") multiplies ×1.1 on 4.6+ models. Example: Opus 5.5 US-only $4.40 / $22. Do not paste Global into a US-only invoice.

DeepSeek — use api-docs EN / api-docs ZH, not the marketing platform page (those numbers disagree). Current flagship IDs: deepseek-flash, deepseek-v4-pro. Do not treat retired V3/R1 sticker prices as “current.”

EN Peak / Off-peak (USD / MTok):

Model	Window	Cache hit in	Cache miss in	Output
deepseek-flash	Off-peak	$0.003	$0.15	$0.60
deepseek-flash	Peak	$0.006	$0.30	$1.20
deepseek-v4-pro	Off-peak	$0.022	$0.66	$1.98
deepseek-v4-pro	Peak	$0.044	$1.32	$3.96

CN mirror (¥ / MTok): Flash idle ¥0.02 / ¥1 / ¥4, peak ¥0.04 / ¥2 / ¥8; V4-Pro idle ¥0.15 / ¥4.5 / ¥13.5, peak ¥0.30 / ¥9 / ¥27. Peak hours: Beijing Mon–Fri 09:00–12:00 and 14:00–18:00 excluding PRC holidays (EN docs list the matching UTC bands). Search/Fetch: ¥5 or $0.75 per 1k successful calls, billing starts 2026-11-01 00:00 Beijing.

Rough read if you only care about list prices on the USD track: Kimi K3.1 Pro intl ($2/$10) sits near Claude Sonnet 5.5 on sticker; Highspeed ($4/$20) sits on Opus 5.5’s sticker. DeepSeek Flash Peak is an order of magnitude cheaper on output and a different product bet. That is price geometry, not a quality ranking — Moonshot’s own benchmark posts are self-reported; I am not reprinting scoreboards as gospel.

For how DeepSeek’s three OpenCode price paths get mixed up in the wild, see our earlier note: OpenCode DeepSeek V4.1 Flash: don’t mix the three prices.

How I would choose without lying to the spreadsheet

• Need the open-weight flagship on a meter → pay the K3.1 Pro API row on the right regional console; China users: real ¥10+ recharge first.
• Need 1M context or Highspeed inside Kimi Code → budget for Pro+ membership, not Plus, and accept that absolute monthly tokens are unpublished.
• Need Claude’s stack (Batch, prompt caching multipliers, US-only) → Claude’s own table; remember ×1.1.
• Need cheapest high-volume completion and can live with Peak/Off-peak → DeepSeek Flash on api-docs, not a screenshot from a marketing page.
• Waiting on Qwen 4 list prices → wait. Official side has not published a number I can cite.

FAQ

Is Kimi K3.1 Pro API pricing the same in China and internationally?

No. China publishes yuan on moonshot.cn; international publishes dollars on moonshot.ai. Same model family, separate invoices. Do not FX-convert and call it official.

Can I try Kimi K3.1 Pro with the ¥15 new-user voucher in China?

No. Official quickstart and FAQ: the ¥15 voucher is not valid for K3.1 Pro. Minimum paid top-up is ¥10 to unlock.

Does Plus membership include unlimited K3.1 Pro tokens?

Not disclosed as an absolute number. Plus+ can call ordinary K3.1 Pro / K3 in Code; 1M and Highspeed need Pro+. Monthly absolute message/token quotas are not published — only relative multiples and a shared pool.

Is Highspeed twice the price?

On the open platform, Highspeed input and output are double the normal Pro; cache-hit stays ¥1.30 / $0.20. Code docs also describe Highspeed consumption as about 2×.

Should I use DeepSeek’s marketing page for quotes?

No. Use api-docs. Current models to quote: deepseek-flash and deepseek-v4-pro, with Peak/Off-peak. V3/R1 are not the current pricing-page leads.

What about Claude US-only?

Global list first; US-only is ×1.1 on eligible models. Cache writes and hits are separate line items.

More on this: AI subscription & API price comparison, OpenCode DeepSeek three prices, DeepSeek category.

Prices move. Re-check the official URLs above before you lock a budget. Checked 2026-10-24 Asia/Shanghai.