GPT
官方价 1:11:1 official price支持协议Protocols
/v1/chat/completions/v1/responses| 模型名Model name |
|---|
gpt-5.6-terra |
gpt-6-sol |
gpt-6.1-sol |
gpt-6-astra |
帮助中心Help Center
管饱 · 帮助中心管饱 · Help Center
把管饱接进你的应用只需要三步：注册、创建 API Key、改 Base URL。全站按官方 USD 价目 1:1 计费，充值 ¥1 到账 $1，按 token 实时扣费。
Getting started with 管饱 takes three steps: sign up, create an API key, and point your base URL at us. Billing follows official USD pricing at 1:1 — ¥1 buys $1 — charged by tokens in real time.
- 注册即用 · 充值 ¥1 起Sign up in seconds · top up from ¥1
- 任意邮箱注册，无需邀请码Any email, no invite code
- ¥1 = $1¥1 = $1
- 支付宝充值Alipay top-up
- 按量计费，不卖包月Pay as you go, no monthly plan
快速开始Quick start
三步接入，改一个 Base URL 就能发出第一条请求，不需要安装额外组件。Three steps and one base URL are all it takes to send your first request. No extra components to install.
-
注册账号Create an account
打开 directapi.cn 用任意邮箱注册 —— 注册完全开放，无需邀请码。注册后到「充值」页充值即可开始调用（¥1 = $1，最低 ¥1），按实际 token 用量扣费，不卖包月。
Open directapi.cn and sign up with any email address — registration is fully open, no invite code needed. Then top up from the Top-up page (¥1 = $1, ¥1 minimum) to start calling; you are billed only for the tokens you use, with no monthly plans.
-
创建 API KeyCreate an API key
登录后进入「API 密钥」页面，新建密钥并选择分组（GPT 组 / Claude 组），复制以
sk-开头的密钥。请创建完立即保存，离开页面后可能无法再次查看完整密钥。After signing in, open "API Keys", create a key and pick a group (GPT or Claude), then copy the
sk-token. Save it immediately — the full key may not be viewable again after you leave the page. -
调用Send a request
把客户端的 Base URL 改成
https://directapi.cn/v1，填入密钥，model使用「支持的模型」里的名称，即可发出请求。Set your client's base URL to
https://directapi.cn/v1, add the key, setmodelto a name from "Supported models", and you are ready to call.
支持的模型Supported models
站内当前在售 19 个模型，分 GPT、Claude、Gemini / Grok、国产四个分组，全部按官方 USD 价目 1:1 计费；model 字段必须与下表逐字一致（大小写、短横线与点号都照抄）。最新清单以站内「模型广场」与 GET /v1/models 为准。
There are 19 models on sale right now, split into GPT, Claude, Gemini / Grok and domestic groups — all billed 1:1 against official USD pricing. The model value must match the table character for character — same case, same hyphens, same dots. The in-app model list and GET /v1/models are always authoritative.
Claude
官方价 1:11:1 official price支持协议（Claude 兼容）Protocols (Claude-compatible)
/v1/messages| 模型名Model name |
|---|
claude-fable-5-1 |
claude-opus-5 |
claude-opus-5-5 |
claude-sonnet-5 |
claude-sonnet-5-5 |
claude-fable-5 |
Gemini / Grok
官方价 1:11:1 official price支持协议（OpenAI 兼容）Protocols (OpenAI-compatible)
/v1/chat/completions/v1/responses| 模型名Model name |
|---|
gemini-3.8-flash |
gemini-4-argon |
grok-4.6 |
grok-4.7 |
国产Domestic
官方价 1:11:1 official price支持协议（OpenAI 兼容）Protocols (OpenAI-compatible)
/v1/chat/completions/v1/responses| 模型名Model name |
|---|
deepseek-v4.1-flash |
kimi-k3 |
kimi-k3.1 |
glm-5.3 |
glm-5.3-flash |
说明 四个分组都按官方 USD 价目 1:1 计费，国产模型组同价、不再打折；Claude 1h 缓存写入按 5m 同价（低于官方 1h 价）。具体上架情况以站内模型广场为准。
Note All four groups are billed 1:1 against official USD pricing; the domestic group is priced the same, with no discount. Claude 1h cache writes are billed at the 5m price (below the official 1h price). Check the in-app model list for what is live.
接口调用API usage
一个 Base URL 同时覆盖 OpenAI 兼容与 Claude 兼容协议，密钥统一用 sk- 开头的那一把。
One base URL covers both OpenAI-compatible and Claude-compatible protocols; the same sk- key works for both.
Base URLBase URL
https://directapi.cn/v1
OpenAI 兼容客户端与 SDK 都填这个地址；Claude 兼容请求走 /v1/messages。所有请求都需要带上 Authorization: Bearer sk-xxx（Claude 客户端也可用 x-api-key）。
Use this address for any OpenAI-compatible client or SDK. Claude-compatible requests go to /v1/messages. Every request needs Authorization: Bearer sk-xxx (Claude clients may use x-api-key instead).
| 用途Purpose | 方法Method | 路径Path |
|---|---|---|
| 对话（OpenAI 兼容）Chat (OpenAI-compatible) | POST | /v1/ |
| Responses（OpenAI 兼容）Responses (OpenAI-compatible) | POST | /v1/responses |
| 消息（Claude 兼容）Messages (Claude-compatible) | POST | /v1/messages |
| 模型列表Model list | GET | /v1/models |
cURL · OpenAI 兼容cURL · OpenAI-compatible
curl https://directapi.cn/v1/chat/completions \
-H "Authorization: Bearer sk-xxx" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-sol",
"messages": [
{"role": "user", "content": "Hello from 管饱"}
]
}'cURL · Claude 兼容cURL · Claude-compatible
curl https://directapi.cn/v1/messages \
-H "x-api-key: sk-xxx" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-5",
"max_tokens": 256,
"messages": [
{"role": "user", "content": "Hello from 管饱"}
]
}'Python · openai SDKPython · openai SDK
from openai import OpenAI
client = OpenAI(
api_key="sk-xxx",
base_url="https://directapi.cn/v1",
)
resp = client.chat.completions.create(
model="gpt-6-sol",
messages=[{"role": "user", "content": "你好，用一句话介绍你自己"}],
)
print(resp.choices[0].message.content)需要 openai 1.x 及以上版本；只改 base_url 与 api_key，其余代码不用动。密钥请放在服务端环境变量里，不要写进前端代码或提交到代码仓库。
Requires openai 1.x or newer. Only base_url and api_key change — nothing else in your code does. Keep keys in server-side environment variables; never ship them in front-end code or commit them to a repository.
计费与充值Billing and top-up
按 token 用量扣费，充值即用。站点不卖包月，也没有套餐门槛。You pay for tokens as you use them. There are no monthly plans and no package minimums.
按 token 计费Charged by token
输入与输出 token 分别按站内价目单价计费，请求成功后结算；流式请求在流结束时结算。
Input and output tokens are priced separately from the in-app rate table and settled when the request completes; streaming requests settle when the stream ends.
余额实时扣减Real-time balance
每次成功调用后余额立即更新，仪表盘与「用量」页面都能看到明细。余额不足时请求会被拒绝，充值后立即恢复。
Your balance updates right after every successful call; details are visible on the dashboard and the Usage page. Requests are rejected once the balance runs out and resume immediately after a top-up.
支付宝充值Top up with Alipay
充值 ¥1 到账 $1，单笔最低 ¥1；支付成功后额度自动入账，可在订单页核对状态。
Every ¥1 tops up $1, with a ¥1 minimum per order. Credit is added automatically after payment succeeds; check the order page for status.
订单 30 分钟过期Orders expire in 30 minutes
未支付的订单会在 30 分钟内自动过期；过期后重新下单即可，请不要对同一笔订单重复付款。
Unpaid orders expire automatically after 30 minutes. Place a new order instead, and never pay the same order twice.
定价口径Pricing policy
站点按官方 USD 价目 1:1 计费（官方标 $5/1M 就是 $5/1M，不加倍率）；充值 ¥1 = $1；四个分组同价，国产模型组不再打折；Claude 1h 缓存写入按 5m 同价；不卖包月，全部按量扣费。
Billing is 1:1 with official USD pricing — an official $5/1M stays $5/1M, with no multiplier on top. Top-ups are ¥1 = $1. All four groups use the same official prices; the domestic group is no longer discounted; Claude 1h cache writes are billed at the 5m price. No monthly plans: everything is pay-as-you-go.
对账与退款：订单状态、扣费金额有疑问，请把订单号、发生时间、模型名和（如果响应头里有）请求 ID 发到客服邮箱，我们核对后答复；需要退款也走同一个入口，由客服人工处理。
Reconciliation and refunds: if an order or a charge looks wrong, email support with the order number, timestamp, model name and (if present in the response headers) the request ID. We will check the records and reply. Refund requests go through the same channel and are handled by support.
常见问题FAQ
调用报错、账号、发票这些高频问题都在这；没找到答案就发邮件给客服。The usual suspects — errors, accounts, invoices. If your question is not here, email support.
01模型名到底该怎么写？How exactly should model names be written?
必须与站内模型名完全一致，直接复制即可。Claude 的版本号用短横线：claude-opus-5-5 可以，写成 claude-opus-5.5（点号）会返回 model not found；而 gpt-6.1-sol、gpt-5.6-terra 里的点号本身就是名称的一部分，不要改成短横线。拿不准时先请求 GET /v1/models 拿准确名字。
Copy the name exactly as listed. Claude versions use hyphens: claude-opus-5-5 works, while claude-opus-5.5 returns model not found. In gpt-6.1-sol and gpt-5.6-terra the dot is part of the name — do not replace it with a hyphen. When in doubt, call GET /v1/models and use the returned name.
02返回 401 / 403 怎么办？What do 401 and 403 mean?
401 表示没带密钥或密钥无效：检查 Authorization: Bearer sk-xxx 是否完整、有没有多余空格或换行，密钥是否已被删除或禁用。403 表示密钥有效但没有权限：通常是这个 Key 的分组不包含该模型，换用对应分组的 Key，或到「API 密钥」确认分组设置。
401 means the key is missing or not valid: check that Authorization: Bearer sk-xxx is complete, free of stray spaces or line breaks, and that the key has not been deleted or disabled. 403 means the key is valid but not permitted: usually its group does not include that model — use a key from the matching group, or check the group on the API Keys page.
03429 是什么？怎么处理？What is a 429 and how do I handle it?
429 表示触发了限速。每个账号默认 240 RPM（每分钟请求数），并发过高或短时间集中请求都会命中。处理方式：降低并发、在请求之间加间隔、对 429 做指数退避重试。业务量确实更大时，发邮件说明场景。
429 means you hit the rate limit. Each account defaults to 240 RPM (requests per minute); high concurrency or bursts will trip it. Lower your concurrency, space requests out, and retry 429s with exponential backoff. If your workload genuinely needs more, email us with the details.
04常见错误码对照Common error codes
| 状态码Status | 含义与处理Meaning and fix |
|---|---|
400 | 请求格式或参数不对，常见是 model 名写错、JSON 不合法、缺少必填字段Malformed request or parameters — usually a wrong model name, invalid JSON, or a missing required field |
401 | 未携带密钥或密钥无效Missing or invalid API key |
403 | 密钥无权使用该分组 / 模型，或已被禁用Key not allowed for that group or model, or the key is disabled |
404 | 模型不存在（model not found），多为名称不一致Model not found — usually a name mismatch |
429 | 触发限速（RPM / 并发）Rate limited (RPM or concurrency) |
500 / 502 / 503 | 网关或上游异常，稍后重试；持续出现请联系客服Gateway or upstream error — retry later; contact support if it persists |
05怎么查看用量和余额？How do I check usage and balance?
登录后打开「用量」页面，可以按时间、模型、密钥查看请求量与 token 消耗；余额在仪表盘和充值页都能看到。需要逐条对账时，把时间段和密钥名称发给我们。
Open the Usage page after signing in to see request counts and token spend by time range, model and key; your balance is shown on the dashboard and the top-up page. For line-by-line reconciliation, send us the date range and key name.
06忘记密码怎么办？I forgot my password — what now?
站点暂未开放自助找回密码。请用注册邮箱发邮件到 [email protected]，说明需要重置密码并附上注册邮箱；我们核对后为你重置，并回复临时密码。
Self-service password reset is not enabled yet. Email [email protected] from your registered address, say that you need a reset, and we will verify the account and reply with a temporary password.
07可以开发票 / 走对公吗？Can I get an invoice or pay by bank transfer?
线上充值目前只支持支付宝。需要发票或对公转账的，请邮件说明抬头、金额和用途，客服核对到账记录后按实际金额处理；能否开具、如何开具以客服答复为准。
Online top-ups currently run through Alipay only. For invoices or corporate bank transfers, email us the invoice title, amount and purpose. Support will check the payment records and handle it based on the amount actually received; whether an invoice can be issued is confirmed by support.
08支持流式输出吗？Is streaming supported?
支持。OpenAI 兼容接口传 "stream": true 就会返回 SSE 增量；Claude 的 /v1/messages 同样支持 stream。如果连接中途断开，客户端重新发起请求即可。
Yes. Set "stream": true on OpenAI-compatible endpoints to receive SSE chunks; /v1/messages supports stream as well. If the connection drops mid-stream, just issue the request again from your client.
09额度、并发和限速是怎么算的？How do quota, concurrency and rate limits work?
计费只看实际 token 用量，不卖包月，也没有套餐门槛。每个用户默认 240 RPM；余额不足时请求会被拒绝，充值后立即恢复。超出限速会收到 429，按上一条的方式退避重试。
You are billed only for the tokens you actually use — no monthly plans, no package tiers. Each user gets 240 RPM by default. Requests are rejected when the balance runs out and resume as soon as you top up. Exceeding the rate limit returns 429; back off as described above.
10充值后多久到账？没到账怎么办？How fast does a top-up arrive? What if it does not?
支付宝支付成功后额度一般会自动入账，可在订单页确认状态。如果长时间仍显示未支付或余额没变，请把订单号和支付时间发到客服邮箱，我们核对后处理；不要重复付款。
Credit normally lands automatically once Alipay confirms payment; check the order page for its status. If the order still shows unpaid or the balance has not changed, email support with the order number and payment time and we will look into it. Do not pay twice.
联系客服Contact support
邮箱（也是唯一官方支持入口）Email — our official support channel
[email protected]工作日 24 小时内回复。Replies within 24 hours on business days.
为了更快定位，来信请附：To speed things up, include:
- 注册邮箱与账号Your registered email and account
- 订单号 / 请求时间 / 模型名Order number, request time, model name
- 错误响应（去掉密钥后再贴）The error response, with keys removed
请不要在邮件里发送完整的 API Key。Please never send a full API key by email.
