free guide tools-and-cost
Choose a ChatGPT or Claude Model
Match current ChatGPT and Claude models to quick, everyday, and longer jobs.
Choose a ChatGPT or Claude Model
Pick a model by the job in front of you, then switch if it struggles.
What this helps you do
Pick a model for writing, coding, or a longer job, with one concrete example for each.
Before you start
Open your ChatGPT or Claude model picker. Available names and limits depend on your plan and may change. Use a short sample task to compare outputs.
Terms to know
A model is the AI system doing the work. Effort controls how much thinking budget it uses for a response. An app can offer several models.
Quick examples
| Model | Good starting job | Coding example |
|---|---|---|
| Luna | Summarize a call or sort comments into topics. | Change the label on one button. |
| Sol | Draft an email or compare two tools. | Build a simple form and check that it submits. |
| GPT-6.1 Sol | Work through a larger task that uses files and tools. | Add a signup flow across several files, run tests, and open it in a browser. |
| Astra | Resolve a difficult problem with several possible causes. | Trace a checkout failure through the browser, code, and database before planning the fix. |
In regular ChatGPT, GPT-6 uses Sol for Instant through Extra High and GPT-6 Pro uses Astra. Luna and GPT-6.1 Sol are model choices in ChatGPT Work and Codex. The options shown depend on your account. See OpenAI's ChatGPT model guide and Work and Codex guide.
| Claude model | Good starting job | Coding example |
|---|---|---|
| Haiku | Sort messages, make short summaries, or handle many repeat tasks. | Rename a button or make another narrow edit. |
| Sonnet | Write a first draft or finish a well-scoped task quickly. | Build a signup form or fix a bug when you know the broken screen. |
| Opus | Plan or complete serious work with several steps. | Find why a payment succeeds but its confirmation email never sends, then fix and test it. |
| Fable | Take on the hardest long-running project when Opus at higher effort falls short. | Move an app to a new database across many files and test runs. |
Anthropic's model overview recommends Opus for most work and Fable for demanding, longer work that still falls short on Opus. Haiku favors speed. Sonnet balances speed and capability.
Step-by-step
- Name the job and what a correct result would look like. A small edit and a full checkout repair are different jobs.
- Pick the row above that matches the job. Check which models appear in your app's picker.
- Give that model the task, files, and success check.
- If it misses the check, give the same brief to the next model you are considering. Compare the finished result, time, and amount of editing you needed.
- Keep the model that completes the job with the least unnecessary time and usage.
Copy this
Paste this in the tool named above. Replace each bracketed detail with your own information.
Copy this
Task: [TASK]. Success means [CHECKABLE RESULT]. Constraints: [TIME, LENGTH, SOURCES, OR TOOLS]. Ask for only the information needed to start, complete the task, and show me the result I should verify. I will run this same brief on two models and compare the output.Things to know
Model names, plan access, and defaults change. Check OpenAI’s current ChatGPT model guide and Claude model settings. A more capable model is not automatically the best for a small task.
If something goes wrong
If the preferred model is missing, check your plan and current picker. If a long response hides the answer, ask for the result first and a short rationale after it.
Final check
A pass is choosing a model for one specific task and confirming the output meets a checkable result.