Reference · Cheatsheet
Token & cost quick reference
Rough math for planning prompts, RAG, and API budgets.
Rough math for planning prompts, RAG, and API budgets.
Rules of thumb
| Measure | Approximate |
|---------|-------------|
| 1 token | ~4 characters English |
| 100 tokens | ~75 words |
| 1 page text | ~500–700 tokens |
| 10-page doc | ~5,000–7,000 tokens |
Estimate one chat turn
input_tokens = your prompt + system + retrieved chunks + history
output_tokens = model reply (capped by max_tokens)
cost = (input × input_rate) + (output × output_rate)Save tokens
- Trim chat history — summarize old turns
- Retrieve **top 3–5** RAG chunks, not 20
- Put stable instructions in system prompt once, not every user message
- Use smaller models for classification/routing
Example
2,000 input + 400 output at $0.15/M in + $0.60/M out:
→ (2000×0.15 + 400×0.60) / 1,000,000 ≈ **$0.00054** per turn