Tinker Reasoning models & thinking budgets

by Kondasamy Jayaraman · drawn with Colorfigures by April Zhu

...

The desk: the model, an effort dial set to none, and a clock.

Thinking recorded

Run a question to see what the model wrote before it answered.

Chapter 1 of 7

Answer now or think first

Larkspur’s assistant reads the right documents and passes the Evals module’s release gate on policy questions. Then sales starts sending it quotes: volume prices, split order lines, a few sums in a row. The answers come back fast, and some are wrong.

This chapter: What changes when a model writes out its working before it answers?

You will learn

  • What a reasoning model’s thinking is: text it writes before the answer
  • Why a question with several steps goes wrong when answered at once
  • Why the same model, with the same documents, can give a different answer after thinking
Words you’ll meet (3)
Reasoning model
A model trained to write out its working before it gives an answer.
Thinking
The working a reasoning model writes before its answer. Readers of the reply usually see a summary of it, or nothing.
Multi-step question
A question whose answer needs several small steps in a row, like picking a price and then multiplying.

The picture on the left is live. Each step below asks you to do one thing in it; the story continues here as you go.

Step 1 of 4

A salesperson forwards a customer’s question to Larkspur’s assistant. It reads the price list and answers at once. The answers in this module were written ahead of time for this site: they are recorded, not live. The grade under each runs in your browser, with the key-fact check from the Evals module.

The question

“A customer orders 55 Kestrel K2 scanners as two order lines, one of 30 and one of 25. How much more do they pay than if all 55 were on one line?”

What the assistant reads: Price list, 2025 (1-9 units / 10-49 units / 50+ units): Kestrel K2 handheld scanner $480 / $445 / $410. Volume prices apply per order line.