instibrain
instibrain  /  guides  /  ai hygiene  /  lesson 6
lesson 6 of 8 · ai hygiene

The smallest model that does the job.

Pick the model and the effort for the task. A small task on the biggest model, thinking at length, pays for work it did not need.

new words

Model
One version of Claude. Opus is the biggest, Sonnet the middle, Haiku the smallest and fastest.
Effort
How long the model thinks before it answers, from low to max.
Thinking tokens
What the model writes to itself before its answer. They are billed as output tokens.

the lesson

Anthropic's list prices per million tokens, read and written:

ModelReadWrite
Opus 5.5$4$20
Sonnet 5.5$2$10
Haiku 5.5, prompts up to 100,000 tokens$0.10$0.50
Haiku 5.5, prompts over 100,000 tokens$0.50$2.50

For prompts up to 100,000 tokens, Haiku costs 40 times less than Opus: $4 / $0.10. Over 100,000 tokens, Haiku's prices rise, and it costs 8 times less: $4 / $0.50. Sonnet costs half of Opus.

Effort costs too. Thinking tokens are billed as output, so a 2,000-token answer after 10,000 tokens of thinking bills 12,000:

12,000 × $20 / 1,000,000 = $0.24 on Opus 5.5
 2,000 × $20 / 1,000,000 = $0.04 for the answer alone

On Opus 5.5 the thinking of earlier turns also stays in the window and is read again, as input, with every later message. On a Claude plan, Anthropic names the model and the effort level among what uses up your limits.

So match them to the job: a rename or a summary needs a small model at low effort, a plan or a hard bug the big one thinking long. Set both when the window opens, not halfway: a change misses the cache of lesson 3. In Claude Code, /model chooses the model, and effort is set as Claude basics, lesson 3 shows; instibrain's setup starts you on Opus at xhigh.

The read price of three models, drawn to scale, from Anthropic's price list. Haiku's price depends on the length of the prompt.

try it

Type this and press Enter:

Which model and effort would you pick to rename 20 files by a pattern, and which to plan the move of our invoices to a new system? Say why for each.

you should see

  • A small model, or low effort, for the rename: the task is mechanical.
  • A big model at high effort for the plan: it needs judgment over many steps.