Billing and usage | Tutujin API Docs

Tutujin API charges for actual usage: your balance is in CNY, with no monthly fee or minimum spend. Most models are billed per token, with input and output priced separately; some, such as image generation models, are billed per request. Every request's model, group, tokens and cost are recorded under Usage in the console.

Two billing methods

Method	Models	How it is calculated
Per token	Most chat, embedding and rerank models	Input tokens × input price + output tokens × output price; prices are quoted per 1M tokens
Per request	Some image generation, audio and other models	A fixed price per request

Which method a model uses and what it costs is shown in the live price list under Models and pricing. The same model costs different amounts in different groups — see Choosing a group.

Token billing in detail

Item	What it covers
Input	System prompt, chat history, documents you pass in, tool definitions and images converted to tokens
Output	Everything the model generates. A reasoning model's thinking counts as output tokens
Cache read	The part of the input served from the context cache, billed at the cache-read price, usually far below the normal input price
Cache write	Some models charge separately for writing content into the cache, at the cache-write price

The cost of one request:

Text

cost = uncached input tokens ÷ 1,000,000 × input price
     + cached input tokens ÷ 1,000,000 × cache-read price
     + output tokens ÷ 1,000,000 × output price

For example, a support bot sends 2,000 tokens of prompt and history and gets a 400-token answer, 50,000 times a month:

Cost estimate · deepseek-v4.1-flash (China Models)

Unit price: input ¥2.8 / output ¥11.2 (per 1M tokens)
Usage per request: 2,000 input tokens, 400 output tokens
Requests: 50,000
Cost per request: ¥0.0101

Total about ¥504Prices in CNY, cheapest group for each model; updated Nov 12, 2026, 8:17 PM.

If the system prompt never changes and the model supports context caching, most of the input can be billed at the cache price. Same volume, assuming 1,500 of the 2,000 input tokens hit the cache:

Cost estimate · deepseek-v4.1-flash (China Models)

Unit price: input ¥2.8 / output ¥11.2 (per 1M tokens)
Usage per request: 2,000 input tokens (1,500 from cache), 400 output tokens
Requests: 50,000
Cost per request: ¥0.00596

Total about ¥298.2Prices in CNY, cheapest group for each model; updated Nov 12, 2026, 8:17 PM.

Each vendor decides when the cache hits; the usage field shows how much actually did. The mechanics are in What context caching is and how it saves money.

Per-request billing

Per-request models charge a fixed amount per call, regardless of prompt length. For example, 1,000 images from a per-request image model:

Cost estimate · nano-banana-pro (Standard)

Unit price: ¥0.4 / request
Requests: 1,000
Cost per request: ¥0.4

Total about ¥400Prices in CNY, cheapest group for each model; updated Nov 12, 2026, 8:17 PM.

Not every image or audio model is billed per request; some charge for input and output tokens instead. The price list is the reference, and Usage lets you check every charge line by line.

Where to see usage

The usage field in each response: Chat Completions returns prompt_tokens and completion_tokens, plus prompt_tokens_details.cached_tokens on a cache hit; Claude Messages returns input_tokens, output_tokens and cache fields.
Usage in the console: every request with time, key, model, group, tokens and cost, with filters and CSV export for reconciliation.

Whether a failed request was charged is shown by the records under Usage.

Balance and spending limits

The balance belongs to the account and all keys share it. When it runs out, requests return 403 with the code insufficient_user_quota.
Each key can have a spending limit; once reached, that key returns 401 and other keys keep working. Use it to cap each project or contractor.
Charges use the price in effect when the request is made.

Invoices

We can issue overseas invoices: add our WeChat support JsnonoChat and send your order number. Chinese domestic invoices can be discussed case by case.

Still stuck? Add our WeChat support JsnonoChat and send the request id from the error message, and we will look into it.

Two billing methods

Method	Models	How it is calculated
Per token	Most chat, embedding and rerank models	Input tokens × input price + output tokens × output price; prices are quoted per 1M tokens
Per request	Some image generation, audio and other models	A fixed price per request

Which method a model uses and what it costs is shown in the live price list under Models and pricing. The same model costs different amounts in different groups — see Choosing a group.

Token billing in detail

Item	What it covers
Input	System prompt, chat history, documents you pass in, tool definitions and images converted to tokens
Output	Everything the model generates. A reasoning model's thinking counts as output tokens
Cache read	The part of the input served from the context cache, billed at the cache-read price, usually far below the normal input price
Cache write	Some models charge separately for writing content into the cache, at the cache-write price

The cost of one request:

Text

cost = uncached input tokens ÷ 1,000,000 × input price
     + cached input tokens ÷ 1,000,000 × cache-read price
     + output tokens ÷ 1,000,000 × output price

For example, a support bot sends 2,000 tokens of prompt and history and gets a 400-token answer, 50,000 times a month:

Cost estimate · deepseek-v4.1-flash (China Models)

Unit price: input ¥2.8 / output ¥11.2 (per 1M tokens)
Usage per request: 2,000 input tokens, 400 output tokens
Requests: 50,000
Cost per request: ¥0.0101

Total about ¥504Prices in CNY, cheapest group for each model; updated Nov 12, 2026, 8:17 PM.

If the system prompt never changes and the model supports context caching, most of the input can be billed at the cache price. Same volume, assuming 1,500 of the 2,000 input tokens hit the cache:

Cost estimate · deepseek-v4.1-flash (China Models)

Unit price: input ¥2.8 / output ¥11.2 (per 1M tokens)
Usage per request: 2,000 input tokens (1,500 from cache), 400 output tokens
Requests: 50,000
Cost per request: ¥0.00596

Total about ¥298.2Prices in CNY, cheapest group for each model; updated Nov 12, 2026, 8:17 PM.

Each vendor decides when the cache hits; the usage field shows how much actually did. The mechanics are in What context caching is and how it saves money.

Per-request billing

Per-request models charge a fixed amount per call, regardless of prompt length. For example, 1,000 images from a per-request image model:

Cost estimate · nano-banana-pro (Standard)

Unit price: ¥0.4 / request
Requests: 1,000
Cost per request: ¥0.4

Total about ¥400Prices in CNY, cheapest group for each model; updated Nov 12, 2026, 8:17 PM.

Not every image or audio model is billed per request; some charge for input and output tokens instead. The price list is the reference, and Usage lets you check every charge line by line.

Where to see usage

The usage field in each response: Chat Completions returns prompt_tokens and completion_tokens, plus prompt_tokens_details.cached_tokens on a cache hit; Claude Messages returns input_tokens, output_tokens and cache fields.
Usage in the console: every request with time, key, model, group, tokens and cost, with filters and CSV export for reconciliation.

Whether a failed request was charged is shown by the records under Usage.

Balance and spending limits

The balance belongs to the account and all keys share it. When it runs out, requests return 403 with the code insufficient_user_quota.
Each key can have a spending limit; once reached, that key returns 401 and other keys keep working. Use it to cap each project or contractor.
Charges use the price in effect when the request is made.

Invoices

We can issue overseas invoices: add our WeChat support JsnonoChat and send your order number. Chinese domestic invoices can be discussed case by case.