Guide
Which model for which job?
Claude and ChatGPT each open a menu of models. What the names mean, where to start, and when to step up, in plain words.
Open a new chat in Claude, and look next to the send button. There’s a name there, like Sonnet 5.5. Click it, and a menu of models opens. ChatGPT has a model menu too, with a slider in it.
Remember Chad, the smartest intern in the world? That menu decides which Chad you get. In Claude, you choose who does the job. In ChatGPT, there’s one Chad, and you choose how long he thinks.
Here’s the short answer. In Claude, start with Sonnet. In ChatGPT, start with GPT-6 on Instant. Step up only when an answer falls short.
Find the menu
- In Claude, click the model name next to the send button, and choose a model. To change how hard it thinks, click Effort.
- In ChatGPT, open the model menu, choose GPT-6, and move the Thinking slider.
What you see depends on your plan. On a free plan, some choices are missing or locked.
Quick or deep
Every choice in both menus trades speed for depth. A quick setting answers in seconds and uses little of your plan’s limit, which is how much you can use before you wait for it to reset. A deep setting thinks longer before it answers, and it uses more of your limit.
Deeper isn’t better for every job. Anthropic’s own advice: “Using Opus on a task Haiku could handle costs you tokens for no gain and slows you down in the process.” Tokens are how the apps count what you use.
Pick a model, then its effort.
- Haiku 5.5Quick answers, summaries, pulling out detailsEvery plan
- Sonnet 5.5Everyday writing, documents, and analysisStart hereEvery plan
- Opus 5.5Hard, open-ended work that needs judgmentPro and up
- Fable 5.1Long, complex jobs it plans itselfPaid plans, extra on Pro
One model, GPT-6. Pick how long it thinks.
- InstantFast answers to everyday questionsStart hereEvery plan
- MediumQuestions that need some thoughtPlus and up
- HighHarder questionsPlus and up
- Extra HighThe hardest questionsPro, Business, Enterprise
- ProDifficult tasks and long-running workPro, Business, Enterprise
Claude has four models
Here’s what Anthropic says each one is for:
- Haiku 5.5 is the quickest. Use it for quick answers, summaries, and pulling details out of a page. It’s on every plan, including Free.
- Sonnet 5.5 is the everyday model, for writing, documents, spreadsheets, and analysis. Anthropic’s advice: “If you’re not sure which model to pick, start here.” It’s on every plan, including Free.
- Opus 5.5 is for hard, open-ended work that needs judgment. Try it on a problem Sonnet struggled with. It starts on the Pro plan.
- Fable 5.1 is Anthropic’s most capable model, built for long, complex jobs: you describe the result you want, and Claude plans the steps. It’s the slowest, and it uses the most of your limit. It isn’t on Free, and on Pro it costs extra.
Each model also has an Effort setting, from Low to Max. The menu marks the level Anthropic recommends for that model as Default, and that’s the right balance for most work. Lower it for quicker answers that use less of your limit. Save the highest level for your hardest problems.
ChatGPT has one model and a slider
ChatGPT’s main model is GPT-6. Instead of choosing between models, you choose how long it thinks on the Thinking slider:
- Instant gives, in OpenAI’s words, “fast responses for everyday questions.” Every plan has it.
- Medium and High think longer, for harder questions. They start on the Plus plan.
- Extra High and Pro think longest. OpenAI describes Pro as being “for difficult tasks and longer-running workflows.” Both need a Pro, Business, or Enterprise plan.
The Free and Go plans run a smaller version of GPT-6, called GPT-6 Luna. The other plans run GPT-6 Sol. The menu says GPT-6 either way.
Pick by job
Most of what a team does at work fits one of these rows. Start with the setting in the row, and step up only when the answer falls short.
| The job | Claude | ChatGPT |
|---|---|---|
| A quick question or a summaryLook up a fact, shorten a long email, pull the dates out of a contract. | ClaudeHaiku 5.5Every plan | ChatGPTInstantEvery plan |
| Everyday writing and analysisDraft an email, tidy a document, find the trend in a spreadsheet. | ClaudeSonnet 5.5Every plan | ChatGPTInstantSlide to Medium if it misses |
| A hard callWeigh a price change, compare two supplier contracts, plan a new process. | ClaudeOpus 5.5Pro and up | ChatGPTHighPlus and up |
| A long job you hand offDescribe the result you want, and let it plan the steps. | ClaudeFable 5.1Paid plans, extra on Pro | ChatGPTProPro, Business, Enterprise |
| A pictureA flyer image, a product mock-up, a logo idea. | ClaudeNot in this appDraws charts and diagrams instead | ChatGPTImagesEvery plan |
Before you step up, check what you gave it. A bigger model can’t guess what you left out. If the answer misses, look at the instructions and the brief first.
Claude or ChatGPT?
For everyday work, both write, summarize, read the files you give them, and answer questions. Both companies also release new models every few weeks, so any ranking of the two goes stale fast.
You’ll notice one difference on the first day: pictures.
So use the one your company already pays for. A work account keeps customer files out of personal accounts, and depending on your plan, your team can share one project and one set of instructions.
If your company pays for neither, try both free plans on the same real job. Keep the one whose first draft needs fewer fixes.
Which Chad you send matters less than what’s on his desk when he starts.
