Tokens & Compute Points
Understanding the concept of Tokens helps you understand AI billing and optimize usage costs.
What is a Token
A Token is the basic unit that AI models use to process text. AI doesn't understand text by "characters" or "words" - instead, it splits text into smaller "Tokens" for processing.
Token Characteristics
- Tokens are not equal to characters, nor equal to words
- A Token may be a complete word, or part of a word
- Token lengths differ across languages
Token Calculation
| Language | Approximate Conversion |
|---|---|
| English | 1 Token ≈ 0.75 words, or 4 characters |
| Chinese | 1 Chinese character ≈ 1.5-2 Tokens |
Examples:
"Hello, world!" → About 4 Tokens
"你好，世界！" → About 8-10 Tokens
Why does Chinese consume more Tokens? Because most AI models are primarily trained on English data, Chinese characters require more Tokens to represent.
Why Tokens Matter
1. Billing Unit
Most AI services charge by Token count:
- Input Tokens: Content you send to AI
- Output Tokens: Content AI sends back to you
Output Tokens are usually more expensive than input Tokens.
2. Context Limits
Each model has a Token limit (context window), determining how much content a conversation can hold.
| Model | Context Window |
|---|---|
| GPT 5.2 | 128,000 Tokens |
| Claude 4.5 Sonnet | 200,000 Tokens |
| GPT 5-mini | 128,000 Tokens |
Chatbox AI Compute Points System
If you use Chatbox AI service, we use "Compute Points" to simplify Token billing.
What are Compute Points
Compute Points are the unified billing unit for Chatbox AI. Each model has its own unique compute-point-per-token ratio — different models consume different amounts of Compute Points per Token. Standard models consume fewer points, while premium models consume more, with the exact rate varying by model.
For example, with the Lite plan (2 million Compute Points): DeepSeek V3 can use ~17.8M Tokens, GPT 5 Mini ~13.2M Tokens.
Plan Quotas
| Plan | Monthly Compute Points | Suitable For |
|---|---|---|
| Lite | 2 million | Light usage |
| Pro | 12 million | Daily usage |
| Pro+ | 30 million | Heavy usage |
Calculation Example
Suppose you send 100 Chinese characters (about 150-200 Tokens), and AI replies with 500 characters (about 750-1000 Tokens), totaling about 1000 Tokens:
Different models consume different amounts of Compute Points, depending on each model's specific rate. For example, standard models like GPT 5-mini consume fewer points, while premium models like GPT 5.2 consume more. See the Model Pricing page for detailed rates.
Tips to Save Tokens
1. Be Precise and Specific
Express your needs clearly with specific requirements and constraints, avoiding vagueness that causes AI to need multiple revisions. Overly simplifying your question is not recommended.
❌ "Help me write an article"
✅ "Write a 500-word article on environmental protection for high school readers, including at least 3 specific statistics"
2. Start New Conversations Promptly
Long conversations accumulate context Tokens. When topics change, starting a new conversation resets the context.
3. Set Appropriate Context Length
Adjust "Maximum Context Messages" in Chatbox settings to reduce history sent to AI.
4. Choose the Right Model
- Use standard models for simple questions (e.g., GPT 5-mini)
- Use premium models for complex questions (e.g., GPT 5.2)
How to Check Token Usage
In Chatbox:
- Token consumption is shown below each message
- When using Chatbox AI, check remaining Compute Points in settings
- BYOK users can view usage in their Provider's dashboard