Token Cost Estimator
Breaks down a prompt or workflow into its token components, calculates expected cost against current Claude pricing, flags context window risks, and proposes concrete optimisations (caching, prompt compression, chunking) ranked by savings potential.
Input: Prompt text or workflow description + target Claude model Output: Token breakdown + cost estimate + context window check + optimisation suggestions
Trigger phrases
- "estimate token cost"
- "how much will this cost"
- "context window too large"
- "reduce Claude API cost"
- "prompt token count"
- "optimise for cost"
Step-by-step workflow
STEP 1 — Identify the target model and input
Ask (or infer from context):
- Which Claude model? (claude-3-5-sonnet, claude-3-opus, claude-3-haiku, etc.)
- Is this a single prompt or a multi-turn workflow?
- What is the expected output length? (short reply / paragraph / full document)
If model is not specified, default to claude-3-5-sonnet-20241022.
STEP 2 — Count input tokens
Break the prompt into components and estimate tokens for each:
| Component | Token estimate rule |
|---|---|
| System prompt | Count words × 1.3 |
| User message | Count words × 1.3 |
| Conversation history | Sum of all previous turns |
| Tool definitions | ~200–500 tokens per tool |
| File/document attachments | Characters ÷ 4 |
| Images | 1000–2000 tokens per image (model-dependent) |
Total input tokens = sum of all components.
STEP 3 — Estimate output tokens
- Short reply (1–3 sentences): ~50–150 tokens
- Paragraph response: ~150–400 tokens
- Structured output (JSON, list): ~200–800 tokens
- Full document/code file: ~500–4000 tokens
- Use user's description or previous outputs to calibrate.
STEP 4 — Check context window limits
| Model | Context window |
|---|---|
| claude-3-5-sonnet | 200,000 tokens |
| claude-3-opus | 200,000 tokens |
| claude-3-haiku | 200,000 tokens |
| claude-3-5-haiku | 200,000 tokens |
Flag if: input + expected output > 80% of context window. Flag as critical if: input alone > 90% of context window.
STEP 5 — Calculate cost
Use current Anthropic pricing (as of 2025):
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| claude-3-5-sonnet | $3.00 | $15.00 |
| claude-3-opus | $15.00 | $75.00 |
| claude-3-haiku | $0.25 | $1.25 |
| claude-3-5-haiku | $0.80 | $4.00 |
Cost = (input_tokens / 1,000,000 × input_price) + (output_tokens / 1,000,000 × output_price)
For workflows with multiple calls, multiply by expected call count.
STEP 6 — Suggest optimisations
Rank suggestions by savings potential:
- Prompt caching — If system prompt > 1024 tokens and reused across calls, enable prompt caching (90% cost reduction on cached portion)
- Model downgrade — If task doesn't require reasoning, suggest haiku (10–40× cheaper)
- Prompt compression — Identify verbose sections that can be shortened without losing meaning
- Output length control — Add explicit instructions to limit response length
- Chunking — Break large documents into smaller calls instead of one massive context
STEP 7 — Present the report
## Token Cost Estimate
**Model**: claude-3-5-sonnet
**Scenario**: [description]
### Token Breakdown
| Component | Tokens |
|-----------|--------|
| System prompt | ~450 |
| User message | ~280 |
| History (3 turns) | ~1,200 |
| **Total input** | **~1,930** |
| Expected output | ~400 |
| **Total per call** | **~2,330** |
### Cost Per Call
- Input: 1,930 tokens × $3.00/1M = $0.0058
- Output: 400 tokens × $15.00/1M = $0.0060
- **Total: ~$0.012 per call**
### At Scale
- 100 calls/day: ~$1.16/day / ~$35/month
- 1,000 calls/day: ~$11.60/day / ~$350/month
### Context Window
✅ 2,330 / 200,000 tokens (1.2%) — well within limits
### Top Optimisations
1. 💰 Enable prompt caching on system prompt → save ~$0.004/call (40%)
2. 🔄 Switch to claude-3-haiku for this task → save ~$0.010/call (85%)
3. ✂️ Compress history summarisation → reduce input by ~400 tokens