Universe Pricing
Universe adopts a pay-as-you-go billing model, charging based on the actual number of tokens consumed by the model. Pay for what you use, with no prepayment or monthly fees. This document details billing rules, model prices and cost optimization strategies to help you plan your usage budget reasonably.
Basic Concepts
What is a Token
Token is the smallest unit for models to process natural language text. It can be a word, a number, a punctuation mark, or a subword.
| Language | Approximate Conversion |
|---|---|
| Chinese | 1 Chinese character ≈ 1-2 tokens (common characters about 1, rare characters about 2) |
| English | 1 word ≈ 1-1.5 tokens (common words about 1, long words about 2) |
| Numbers/Symbols | 1 number or punctuation ≈ 1 token |
Actual token count is subject to the
usagefield returned by the interface. Each API call response includes three fields:prompt_tokens(input token count),completion_tokens(output token count) andtotal_tokens(total token count).
Billing Unit
All model prices are quoted in yuan/million tokens. Billing separately counts input tokens and output tokens, each calculated at corresponding unit prices.
Model Prices
Text Generation Models
| Model | Positioning | Input (Cache Hit) | Input (Cache Miss) | Output |
|---|---|---|---|---|
| Universe 3.0 | Lightweight & Efficient | 0.05 yuan/million tokens | 2 yuan/million tokens | 4 yuan/million tokens |
| Universe 3.0 Pro | Professional Enhanced | 0.1 yuan/million tokens | 9 yuan/million tokens | 18 yuan/million tokens |
| Universe 4.5 | Flagship All-in-One | 0.2 yuan/million tokens | 12 yuan/million tokens | 20 yuan/million tokens |
For capability positioning and applicable scenarios of each model, please refer to Model Products.
Price Comparison Overview
The table below shows cost differences between the three models in different scenarios more intuitively (based on 1 million calls, each with 500 tokens input + 200 tokens output, cache miss):
| Model | Input Cost | Output Cost | Total Cost |
|---|---|---|---|
| Universe 3.0 | 1,000 yuan | 800 yuan | 1,800 yuan |
| Universe 3.0 Pro | 4,500 yuan | 3,600 yuan | 8,100 yuan |
| Universe 4.5 | 6,000 yuan | 4,000 yuan | 10,000 yuan |
These are estimation references. Actual costs depend on your call volume and input/output ratio.
Prompt Cache
What is Prompt Cache
Prompt Cache is an optimization mechanism that automatically reduces repeated call costs. When multiple requests contain the same prefix content (such as fixed System Prompt, long documents, tool definitions, etc.), the system automatically caches the processing results of this content.
The same input content in subsequent requests will hit the cache and be billed at a price much lower than the normal price, without modifying code or interfaces. The system automatically identifies and matches.
Cache Hit Prices
When cache hits, input price is only 1%~2.5% of normal input price, greatly reducing repeated call costs:
| Model | Normal Input Price | Cache Hit Price | Savings Ratio |
|---|---|---|---|
| Universe 3.0 | 2 yuan/million tokens | 0.05 yuan/million tokens | Save about 97.5% |
| Universe 3.0 Pro | 9 yuan/million tokens | 0.1 yuan/million tokens | Save about 98.9% |
| Universe 4.5 | 12 yuan/million tokens | 0.2 yuan/million tokens | Save about 98.3% |
How to Maximize Cache Utilization
| Suggestion | Description |
|---|---|
| Keep System Prompt Consistent | Place fixed System Prompt at the beginning of message list, avoid frequent wording changes that cause cache invalidation. |
| Place Long Documents at Input Front | Put unchanged reference materials, context documents, etc. at the prefix position of input content to improve cache hit rate. |
| Keep Tool Definitions Stable | If using Function Calling, try to keep the JSON Schema of tool definitions unchanged. |
| Avoid Inserting Dynamic Content Before Fixed Content | Cache matches by prefix, dynamic content (such as user input) should be placed after fixed content. |
Cache Billing Example
Assuming using Universe 3.0 Pro, each request contains 800 tokens of fixed System Prompt and 200 tokens of user dynamic input:
| Billing Item | First Request (Cache Miss) | Subsequent Requests (Cache Hit) |
|---|---|---|
| System Prompt (800 tokens) | 800 × 9 / 1,000,000 = 0.0072 yuan | 800 × 0.1 / 1,000,000 = 0.00008 yuan |
| User Input (200 tokens) | 200 × 9 / 1,000,000 = 0.0018 yuan | 200 × 9 / 1,000,000 = 0.0018 yuan |
| Input Total | 0.009 yuan | 0.00188 yuan |
After cache hit, input cost is reduced by about 79%.
Billing Rules
Billing Formula
The cost of each API call consists of two parts: input and output:
Single Call Cost = Input token count (millions) × Input unit price + Output token count (millions) × Output unit price
Where input unit price depends on whether cache is hit.
Billing Examples
Scenario: Using Universe 3.0 Pro for text generation, one request with 1,000 tokens input (cache miss) and 500 tokens output.
Input Cost = 1,000 / 1,000,000 × 9 yuan = 0.009 yuan
Output Cost = 500 / 1,000,000 × 18 yuan = 0.009 yuan
Total Single Call Cost = 0.009 + 0.009 = 0.018 yuan
Scenario: Same model, 5,000 tokens input (of which 4,000 tokens hit cache), 1,000 tokens output.
Cache Hit Input = 4,000 / 1,000,000 × 0.1 yuan = 0.0004 yuan
Cache Miss Input = 1,000 / 1,000,000 × 9 yuan = 0.009 yuan
Output Cost = 1,000 / 1,000,000 × 18 yuan = 0.018 yuan
Total Single Call Cost = 0.0004 + 0.009 + 0.018 = 0.0274 yuan
Deduction Method
- Costs are deducted in real-time from account recharge balance, settled immediately after each API call completes.
- If balance is insufficient, API requests will return errors. Please recharge in time to maintain service availability.
- Failed requests (returning non-2xx status codes) do not generate token consumption and are not billed.
Cost Optimization Suggestions
| Optimization Strategy | Applicable Scenario | Expected Effect |
|---|---|---|
| Choose Appropriate Model | Use Universe 3.0 for simple tasks, only use 4.5 for complex tasks. | For the same task, 3.0 costs only about 1/5 of 4.5. |
| Utilize Prompt Cache | Keep System Prompt and fixed prefix consistent. | Cache hits can save about 97%-99% of input costs. |
| Streamline Input Content | Remove unnecessary context and redundant historical dialogues. | Directly reduces input token count. |
| Control Output Length | Set reasonable max_tokens, constrain output length in Prompt. | Avoids unnecessary costs from overly long outputs. |
| Batch Processing | Combine multiple simple tasks into one request. | Reduces request count, improves token utilization. |
Cost Estimation Tool
You can use the following formula to quickly estimate monthly costs:
Monthly Cost ≈ Daily average calls × 30 × (Average input tokens / 1,000,000 × Input unit price + Average output tokens / 1,000,000 × Output unit price)
Tip: It is recommended to check actual token consumption details in the console's cost center, and combine with historical data for more accurate cost estimation.
Frequently Asked Questions
Q: Is there a free trial quota?
New registered users can receive a certain amount of free experience quota. Specific quota and validity period are subject to the instructions on the console registration page. After the free quota is used up, you need to recharge to continue using.
Q: How to view consumption details?
Log in to the console, in the Cost Center you can view daily/monthly consumption trends, token consumption details by model dimension, and independent consumption statistics for each API Key.
Q: Will prices change?
Product prices may be adjusted with market and technology development. If there are changes, we will notify in advance through platform announcements. Please refer to the latest price information on this page and check regularly.
Q: What payment methods are supported?
Supports mainstream recharge methods such as Alipay, WeChat Pay, bank transfer. Enterprise users needing corporate transfers or invoices, please contact the business team.
Q: Will failed requests be billed?
No. Only requests that successfully return results (HTTP 2xx) will generate token consumption and billing. Request failures due to authentication failure, rate limiting, server errors and other reasons do not incur charges.
More Help
- To understand capability differences and selection recommendations for each model, please refer to Model Products.
- To view API call methods and code examples, please refer to API Reference.
- To learn how to write efficient Prompts to reduce call costs, please refer to Prompt Engineering Guide.
- For other issues, please check FAQ or contact technical support.