Input Tokens
Content sent to the model, including user content, system instructions, chat history, file-processing results, and tool definitions.
Start with Models and Prices in Model Plaza
Open Model Plaza to filter current models by platform, group, multiplier, or model name and compare KHaiXAPI prices with official provider reference prices.
- Signed out: Review models, pricing methods, KHaiXAPI prices, official reference prices, and multipliers in public groups.
- Signed in: Also review exclusive groups and personal multipliers available to the current account.
- Distinguish price types: Official reference prices are for comparison and are not KHaiXAPI settlement prices.
- Verify actual charges: Check peak rules, caching, reasoning, multimodal items, and other charges against page notices and console request records.
How Model API Costs Are Calculated
KHaiXAPI models, unit prices, pricing units, multipliers, and availability can change. Review current public pricing in Model Plaza and use the rules applicable at request time and console request records to verify actual costs.
Text generation is billed by actual usage, usually based on input and output token counts, so each request can consume a different amount. Image generation is billed by the number of images actually generated. Each image uses the unit price for the request's final 1K, 2K, or 4K resolution tier; when a request returns multiple images, the charges accumulate by the generated image count. Check Model Plaza and request records for the applicable rules.
Base text cost = Input tokens ÷ Pricing unit × Input unit price + Output tokens ÷ Pricing unit × Output unit price
Use the “KHaiXAPI price” in Model Plaza to understand current pricing. The “official reference price” compares public provider pricing and is not the KHaiXAPI settlement price. Multipliers and peak rules may affect actual costs, which should be verified in request records.
How Much Text Equals One Token?
A token is a unit a model uses to process text. It is not equivalent to a fixed number of Chinese characters, words, or characters. Language, punctuation, numbers, code, spaces, and the model's tokenization all affect the result, so there is no permanently accurate conversion table for “one token equals this many Chinese characters.”
In an AI application, input usage usually includes more than the final sentence visible in the input box. The application may also send system instructions, role settings, chat history, file contents, tool definitions, and other context, all of which affect input tokens.
Billing Items That May Appear on Your Bill
Output Tokens
Text, code, or structured results actually generated by the model. Longer answers usually use more output tokens.
Caching and Reasoning
Some models distinguish cache reads, cache writes, or reasoning-related usage. Whether these are billed separately depends on the model's rules.
Multimodal Input and Tools
Images, audio, video, files, web access, or tool calls may use different measurement methods and cannot be calculated directly with the text formula.
What a Billing Multiplier Means
A multiplier usually represents the conversion between a base cost and the actual settlement cost. For example, 0.8x means settlement at 80% of the base cost, while 1.2x means settlement at 120% of the base cost. Check how the “base cost” is defined and whether the displayed price already includes the multiplier in Model Plaza.
| Model Plaza Item | How to Use It | Important Note |
|---|---|---|
| KHaiXAPI input and output prices | Use with actual token usage to understand current cost | Still check peak rules and request records |
| Official reference price and current multiplier | Compare official pricing with KHaiXAPI pricing | The official reference is not the settlement price |
How CNY Recharge Converts to USD Service Credit
Recharge Conversion CNY 1 = USD 1 in usage credit
Payment Method Alipay
This USD value is account credit used by the platform to bill model requests. It is not a live foreign-exchange rate, nor is it US currency that can be withdrawn or transferred. Balance deductions are still settled according to the model, token usage, group multiplier, and other actual billing items.
When recharging, select Alipay and enter the amount in CNY. After payment, usage credit is added at the conversion rate shown above.
Why the Same Question Can Cost Different Amounts
The same final question visible to the user does not mean the complete request sent to the model is identical. The following factors most commonly cause differences.
Longer Chat History
Ongoing conversations may repeatedly send earlier content, causing input tokens to increase with each turn.
Additional Application Instructions
Different applications add system prompts, role settings, knowledge-base excerpts, or tool descriptions.
Different Models and Output
Models have different unit prices, and the length of generated content will not be exactly the same.
Retries and Parallel Requests
Network interruptions, automatic application retries, or multiple simultaneous requests may create several request records.
Verify a Request Record in Five Steps
Review the model, token usage, multiplier, and request status in the console. A separate API key for each application makes the source easier to identify.
- Confirm that the model name actually used matches the application setting.
- Review the request time, status, and input and output tokens.
- Verify the model unit price, pricing unit, and multiplier basis at the time of the request.
- Check for retries, parallel requests, or duplicate requests around the same time.
- Record the request details and to investigate unexplained costs.
How to Reduce Model API Usage Costs
- Choose a model for the task: Simple organization and classification do not always require the most expensive model.
- Control context: Start a new session when appropriate in long conversations to avoid repeatedly sending history that is no longer needed.
- Limit output length: Clearly specify the required scope to reduce lengthy output unrelated to the task.
- Reduce repeated files: Check whether the application attaches the complete document or knowledge-base content again with every request.
- Check automatic retries: Understand the application's concurrency, retry, and background-task settings to avoid unintended duplicate requests.
- Manage keys separately: Create an API key for each application and review usage and balance changes regularly.
What to Do When You Find Unusual Usage
First pause any application or related API key that may continue sending requests. Then record the request time, model, status, token usage, and cost. Failed, timed-out, or automatically retried requests may produce different records; use the status and usage shown in the console.
If you cannot confirm the cause, prepare the request details and for investigation. For specific billing boundaries, also review the Terms of Service and Service-Specific Terms.
Frequently Asked Questions
What does a billing multiplier mean?
A multiplier describes how a model base price relates to the KHaiXAPI price. Model Plaza shows the KHaiXAPI price, official reference price, and current multiplier; the official reference is not the settlement price, and actual charges are recorded in the console.
How is Token billing calculated?
Text requests usually calculate input and output costs separately: divide input tokens by the pricing unit and multiply by the input price, then divide output tokens by the pricing unit and multiply by the output price. Apply multipliers or other billing items according to the actual rules.
How many Chinese characters equal one Token?
There is no fixed conversion. Token counts depend on language, punctuation, numbers, code, tokenization, and model implementation. Refer to the actual request record.
Why can the same question cost different amounts?
An application may also send system instructions, chat history, file contents, or tool definitions. Output length, retry count, and the model actually used may also differ, so the same visible question does not imply identical request usage.
Are failed or interrupted requests always free?
Failed, timed-out, or automatically retried requests may produce different records. Check request status and usage in the console, and with the request details if anything is unclear.
Why does a CNY 1 recharge provide USD 1 in credit?
Recharging CNY 1 provides USD 1 in usage credit, with Alipay supported for payment. This USD value is account credit used to bill model requests, not foreign-currency exchange or withdrawable US dollars.