Billing
Pay for usage, with a record of every call
No monthly fee and no minimum spend commitment. ¥1 buys ¥1 of credit, and every charge is listed on the console's Logs page.
Prices updated Nov 13, 2026, 5:48 PM
What one support chat turn costs
gpt-6.1-sol · Standard group
- Input (not cached)1,000 × ¥4 ÷ 1,000,000
- ¥0.004
- Input (cache read)2,000 × ¥0.2 ÷ 1,000,000
- ¥0.0004
- Output500 × ¥20 ÷ 1,000,000
- ¥0.01
At 10,000 calls a month, that is about ¥144. Actual charges are whatever the console's Logs page shows.
¥0
monthly fee, and no minimum spend
1:1
¥1 buys ¥1 of credit
739
models, every unit price public
18
groups, one chosen per key
Two ways to bill, one formula
Chat models are billed per token; image generation and similar models per request.
input tokens × input price+output tokens × output price+cache read/write cost=cost of one call
Prices are per 1M tokens. For per-request models, cost = unit price × number of requests.
Per token
Chat, embedding and similar models. Input (what you send) and output (what the model generates) are priced separately, in ¥ per 1M tokens.
gpt-6.1-solinput ¥4 · output ¥20 per 1M tokens
Per request
Image generation and similar models charge once per request regardless of tokens, in ¥ per request.
gpt-image-2¥0.12 per request · 100 requests ¥12
Each call is charged at the price in effect when it is made; price changes only affect later calls.
Cache hits make input far cheaper
Input tokens that hit the context cache are billed at the cache-read price.
−95%cheaper than normal input when read from cache
Example: gpt-6.1-sol in the Standard group, in ¥ per 1M tokens.
- Cache reads
- Input tokens that hit the context cache are billed at the cache-read price, which is usually far below the normal input price.
- Cache writes
- Some models charge a cache-write price when content is written to the cache. Each model's detail page shows whether it has cache prices and what they are.
When caching pays off
- A long, fixed system prompt
- The same document referenced again and again
- Long multi-turn conversations
Same model, different groups, different prices
Each group is a different route with its own source. When you create an API key you pick a group for it, or several groups in priority order.
Price in each group · gpt-6.1-sol
input / output / cache read, ¥ per 1M tokens
Customer-facing, source matters
Prefer groups labelled Official or Cloud.
Internal tools and offline batch jobs
If budget comes first, the cheaper Mixed groups are an option; they include non-official sources.
Chinese models
Use the China Models group for DeepSeek, Qwen, Kimi, GLM, MiniMax and others.
Keep workloads apart
Create one key per product or environment, each with its own group and spending cap.
See all 18 groups
- StandardLowest priceMixed sources618 modelsCovers every model at the lowest price; some GPT and Claude routes are non-official
- AzureCloud116 modelsMicrosoft Azure OpenAI — official, well priced, enterprise-grade
- Azure PlusCloud135 modelsPremium Azure OpenAI line with o-series reasoning models
- OpenAI DirectOfficial169 modelsOpenAI's own API
- OpenAI Direct PlusOfficial174 modelsThe fastest, steadiest OpenAI line, incl. o-series and realtime
- Official · EnterpriseOfficial309 modelsEnterprise line at official list prices, sourced from OpenAI and Anthropic, stability first
- Claude DirectOfficial47 modelsAnthropic's own API; new models may have limited concurrency at launch
- Claude List PriceOfficial47 modelsAnthropic official line billed at list price
- Claude PlusMixed sources40 modelsCost-effective Claude line; includes non-official sources, prices may change
- Claude LiteMixed sources27 modelsThe cheapest Claude line, non-official sources
- Gemini PlusOfficial70 modelsGoogle Gemini official line with video understanding
- Gemini High TierOfficial70 modelsHigher-quota Google Gemini official line
- China ModelsLowest price—195 modelsDeepSeek, Qwen, Kimi, GLM, MiniMax and more at special prices
- China Models 2—18 modelsBackup line for Chinese models
- Economy (per token)Mixed sources42 modelsNon-official sources, billed per token at a discount
- Economy (per call)Mixed sources54 modelsNon-official sources, billed per call — good for long prompts
- Stable Diffusion—5 modelsStable Diffusion image models
- Video Line 1Mixed sources6 modelsVeo and Grok video generation; non-official sources
Top up, get charged, reconcile
Top up first, credit is drawn down by actual usage, and every charge is in the console.
01
Top up
Top up online (as shown in the console) or buy redemption codes offline through support.
02
Pay per use
¥1 buys ¥1 of credit. Credit is drawn down by actual usage, with no monthly fee and no minimum spend commitment.
03
Reconcile
The console shows where every yuan went.
- Usage
- time, model, group, token counts and cost of every call
- Per-key caps
- a spending cap and expiry date for each key
- Top-up history
- amount and time of every top-up
Invoices
We issue both China domestic invoices (fapiao) and overseas invoices: add our WeChat support and send your order number.
Check the prices, then choose your groups
Every model and group price is public. For bulk purchases or invoices, add our support on WeChat.
Scan the code or search the WeChat ID to add us