CDP
LIGHTER AI GUIDECDP and CDP Builders
USE LESS ENERGY

Pick the lightest setting that does the job

Most everyday work doesn't need the biggest model or the most thinking. Starting light saves energy and stretches your usage limits too. Here's what to pick for common tasks in Claude and ChatGPT.

WHY IT MATTERS

Thinking is the biggest factor

Thinking costs the most

When a model "thinks" before it answers, it does a lot more work. Research has found that this can use around 30 times more energy than answering right away.

Smaller models use less

Faster, lighter models like Haiku in Claude or Instant in ChatGPT use less computing power for each answer.

Redoing costs too

Every redo or long back and forth is another round of work. A clear first ask usually saves more than any setting.

THE GUIDE

What to pick for each task

In Claude, the dropdown next to the message box sets the model and the Effort. In ChatGPT, the Chat tab has a thinking slider (Instant, Medium, High, and so on), and the Work tab also lets you pick a model.

TASKCLAUDECHATGPT
LIGHTQuick email, reply, or rewriteProofreading, changing the tone, a short answerHaiku 5.5, Effort LowChat tab, Instant
LIGHTSummarize something you paste inMeeting notes, an email thread, a short documentHaiku 5.5 or Sonnet 5.5, Effort LowChat tab, Instant
MEDIUMEveryday writingA memo, letter, agenda, or announcementSonnet 5.5, Effort MediumChat tab, Medium
MEDIUMPull details out of a fileDates and deadlines, a list of contacts, key termsSonnet 5.5, Effort MediumChat tab, Medium
MEDIUMMake a fileA Word document, slides, or a spreadsheetOpus 5.5, Effort Medium (the default)Work tab, Default model, Medium
HEAVYCompare or check numbersTwo bids or proposals side by side, a budget, a spreadsheet reviewOpus 5.5, Effort Medium, or High if it's high stakesWork tab, GPT-6.1 Sol, Medium or High
HEAVYA tricky decision or careful reviewLots of moving parts, contract language, real stakesOpus 5.5, Effort HighWork tab, GPT-6.1 Sol, High
HEAVYResearch across many sourcesFunding programs, regulations, market questionsResearch, only when a normal chat can't answer itDeep research, only when a normal chat can't answer it

Save for the hardest jobs: Fable 5.2 and Effort Extra or Max in Claude, and GPT-6 Astra, Extra High, or Pro in ChatGPT. These are the heaviest options and also use up your limits fastest.

HABITS

Small habits that add up

GOOD TO KNOW

How sure is this?

Neither Anthropic (Claude) nor OpenAI (ChatGPT) publishes energy numbers for its models, so there's no exact chart. This guide follows what outside research consistently shows: answers with thinking turned on use far more energy than quick answers, and smaller models use less than larger ones. Treat it as a rule of thumb, not a measurement. Sources: Hugging Face study on reasoning models and estimates of AI energy use per answer.

Model names are as of November 2026 and will change over time. The idea stays the same: pick the lighter option first.