Usage Credits How They Are Calculated
Learn how input, output, cache tokens, coding agents, multimodal tasks, and group multipliers affect the Credit cost of each call.
Model pricing, usage, and the group multiplier determine the final deduction.
A request may contain three main token types.
Input Token
User questions, system prompts, message history, code, tool results, and retrieved content may all count as input.
Output Token
Model-generated text, code, structured results, and tool arguments usually count as output.
Cache Token
Some upstream providers support cache reads or writes with separate pricing. Refer to the model pricing page and usage logs.
Calculated using prices per 1M tokens.
Assumed Base Model Price
- Input：1 Credit / 1M Token
- Output：4 Credit / 1M Token
- Input this request: 100,000 tokens
- Output this request: 20,000 tokens
Input: 100,000 / 1,000,000 × 1 = 0.1 Credit Output: 20,000 / 1,000,000 × 4 = 0.08 Credit Base usage: 0.1 + 0.08 = 0.18 Credit
The same base usage under different groups.
| Base Usage | Group Multiplier | Final Deduction | Explanation |
|---|---|---|---|
| 0.18 Credit | 0.2× | 0.036 Credit | Base usage multiplied by 0.2 |
| 0.18 Credit | 1× | 0.18 Credit | Same as base usage |
| 0.18 Credit | 2× | 0.36 Credit | Base usage multiplied by 2 |
Credit and the actual RMB purchase cost.
Top-Up Ratio
OmniDino currently uses 1 RMB = 1 USD Credit. Paying ¥100 gives you 100 Credits.
Actual RMB Cost
If a request deducts 0.18 Credit, it consumes about ¥0.18 of purchased balance. A 0.2× group deducts 0.036 Credit, equal to about ¥0.036.
Why do Codex and Claude Code often consume many tokens?
Long Context
Project files, conversation history, system instructions, and tool output may repeatedly enter the context.
Multi-Step Execution
One task may involve multiple model calls for analysis, reading, editing, testing, and review.
Tool Results
Terminal output, code diffs, and error logs increase input tokens and may trigger additional reasoning.
Not every capability is billed by text tokens.
Image Generation
Usually billed by model, size, quality, number of images, or the upstream billing unit.
Audio Models
May be billed by audio duration, characters, requests, or model-specific units.
Embedding / Rerank
Embeddings are often billed by input tokens, while rerank may be billed by tokens, requests, or document count.