Claude Code Tools

Token Cost Prompts: 6 to Run Before You Switch Provider

Sonnet 5 rose to 3 and 15 dollars per million today with a tokenizer change adding 10 to 35 percent more tokens on code, Claude Code weekly limits tighten on 14 September, and the GPT-5.6 Sol promotional rate reverts in November. Before moving a workload, it is usually cheaper to stop wasting tokens on the one you have. These six prompts find the waste.

⌨️ 6 prompts 🕐 Updated Aug 31, 2026

💡 How to use these prompts: Replace everything in [BRACKETS] with your specific details before sending. Click Copy to copy any prompt to your clipboard instantly.

Find what is padding my prompt

System prompts accumulate. Every request pays for the padding, forever.

Here is a system prompt I send on every request: [PASTE]

Tell me:
1. Which instructions are load-bearing and which could be removed with no behaviour change
2. Anything repeated in different words
3. Anything the model would do anyway without being told
4. What this would look like at roughly half the length

Then show the shortened version. Do not remove anything that changes output.

Work out what context is actually needed

Passing full context every step is the single biggest waste in most agent loops.

My agent currently receives this context on every step: [DESCRIBE]

For each item, tell me:
- Whether it is needed at every step or only some
- What breaks if it is omitted
- Whether it could be fetched on demand instead of sent every time

Rank by tokens saved against risk of breaking something.

Route cheap work to a cheap model

Most pipelines run everything on one model. Routing simple work to a cheap tier is the largest saving available.

Here are the tasks in my pipeline: [LIST]

For each, tell me the cheapest model class that could plausibly handle it, and what would fail if I used something cheaper still.

Flag anything where I am using a frontier model for work a small one could do. Be specific about what the failure would look like, not just that quality would drop.

Cap the output before it runs long

Output tokens cost five times input on most providers. Constraining length is the fastest lever.

This prompt produces longer responses than I need: [PASTE]

Rewrite it to constrain output. Specify:
- Maximum length and in what unit
- What to omit entirely
- What to do if the answer genuinely needs more room

Show the rewrite, then tell me what quality risk the constraint introduces.

Model the cost before you migrate

The last line is the important one. A three-month discount rarely covers a migration.

I am considering moving this workload from [MODEL A] to [MODEL B]: [DESCRIBE]

Walk me through:
- Roughly how token volume differs between the two, including tokenizer differences
- What the migration costs in engineering time
- What breaks and needs retesting
- How long the saving takes to repay that cost

Price the destination at its standard rate, not any promotional rate.

Audit a week of spend

Most cost advice is generic. Pointing a model at your own numbers gets a specific answer.

Here is my usage breakdown for the last week: [PASTE]

Tell me:
- Where the spend concentrates
- Whether that matches where the value is
- Which single change would cut the most with the least risk
- What I should measure to know if it worked

Do not give me a list of general optimisations. Tell me what to do first.