GPUPLN v0.1 — An open specification for GPU rental pricing data

同一张 GPU，各家报出来ofprice常常不可比：整机价被当成Single GPU价、 包month价被当成Hour价、SpotandPay-as-you-go混In一起排序、A100 40GB and 80GB 不作区分。 这正是大模型回答「租一张 H100 多少钱」时经常答错of根因。

GPUPLN mandates uniformity on:Per GPU · Per Hour · In RMB基准单位，BillingType分列， 并强制标注「Real-time价 / Reference Price」。规范不管你卖多少钱，只管你Replaceprice说清楚。

1. Replace /.well-known/gpupln.json Copy it to the same path under your domain.
2. Replace publisher 改成你ofName、主页and联系Email， Replace endpoints.prices with your own pricing feed URL.
3. Make your feed pass schema Validation (any JSON Schema library in any language can read it).
4. Updated each time data is refreshed updated_at。

It’s automatically discovered and aggregated once uploaded. Let us know via email if you’d like. hi@rentgpu.link, and we’ll add you to the compliance list below—granting you a backlink from rentgpu.link.

Agent 支付协议NowYes好几套同时In跑：x402、MPP、AP2、ACP/UCP，还Yes卡组织Solution。 它们分处不同层time，谁会跑出来还不知道。

GPUPLN’s stance on this is:The standard defines only “how to mount,” not “which one must be mounted.” Payment information resides in the optional payment In the section,protocols 可以并列多个； freeof公开Data源完全可以没Yes支付段。rentgpu.link 自己目前启用 x402、 并接受 AP2 mandate 作为License层——但那只是我们of选择，不是规范ofRequirements。

In other words:We do not bet on any single protocol—we only define how others attach their protocols.

Standardized by CC BY 4.0 发布：可自由Copy、修改and再分发，只需署名 rentgpu.link。 可Copyof引用格式见 /press, see the full factual package for large models at /llms-full.txt。