Turn a result into a decision

A playground result is only useful once it becomes a sentence: for this task, this model, because of this evidence, at this cost. Write yours, price it for a month, and know when to re-test.

In this lesson you will

Write a one-line model decision with the evidence attached
Price that decision per month at your real volume
Know the three events that should send you back to the playground

Most model choices are never written down, so they are never revisited, so they quietly cost flagship money for years. The fix is a sentence. Fill this one in from the run you did in lesson 3 or a leaderboard row from lesson 2.

Your turn

Your model decision

The task, specificallyThe model you choseThe evidenceFrom the cost line or the leaderboardYour honest volumeCost per run × runs

For [TASK] I will use [MODEL] because [REASON]. It costs about [COST PER RUN] per run, so at [RUNS PER MONTH] runs a month that is roughly [MONTHLY COST]. I will re-test if the model is replaced, the prompt changes, or a failure reaches a customer.

Hint: If the monthly cost is under a dollar, the decision is not worth agonising over; spend the time on the checks instead.

Price it before you believe it

Two real decisions priced at 2,000 runs a month, using per-run costs from the bench of 2026-09-12.
Task	Model	Cost per run	At 2,000 runs / month	Flagship alternative
Support-thread summary	Claude Haiku 4.5	$0.000611	$1.22	Claude Opus 5 at $0.00436 → $8.72, and it failed the check
Stacked-discount arithmetic	Mistral Medium 3.5	$0.00163	$3.26	Ministral 3 8B at $0.000016 → $0.03, and it was wrong

Both rows are small money. That is normal, and it is the reason to make the decision quickly and write it down rather than research it for a week. Put your own volumes into the calculator if the numbers get bigger; the arithmetic is the same.

Knowledge check

Your sentence says: 'For invoice extraction I will use the cheapest model because it is cheapest.' What is missing?

Do it for real

Run your own prompt →Get the cost-per-run line for your task.AI Cost Calculator →Scale it to your monthly volume.First Output 101 →Turn the decision into a script that runs it.

Lesson FAQ

How do I choose which AI model to use for a task?

Run the task on the cheapest model with checks that define success. If it passes, use it and write down the evidence and cost per run. If it fails, move one step up the family and repeat. Re-test when the model is replaced, the prompt changes, or a failure reaches a customer.

How much does it cost to run a model 2,000 times a month?

It depends on the model and the task length. On the bench of 2026-09-12, a 60-word summary cost $0.000611 per run on Claude Haiku 4.5, about $1.22 for 2,000 runs, versus $8.72 on Claude Opus 5. Use the cost line from your own run for your own task.

Should I write this down even for tiny costs?

Yes, because the cost is tiny now and the habit is what protects you when the volume grows. A one-line decision takes a minute and is the difference between re-testing and forgetting.

Price it before you believe it

Lesson FAQ

Stop overpaying for AI tools