The short answer
To choose the right AI model, match it to the stakes: a fast model for low-risk jobs you will read anyway, the main model for judgement work and the top model only for problems the main one fails. In November 2026 that means Claude Haiku or Sonnet 5.5, Opus 5.5 and Fable 5.1 in Claude; GPT-6 Luna, GPT-6 Sol and Pro reasoning in ChatGPT; and Flash, Gemini 4 Argon and higher effort in Gemini. Free plans get the smaller models.
Why the model you pick matters more than the app
Every big AI assistant now offers several models, and they are not interchangeable. In ChatGPT, Claude and Gemini the same prompt can be answered by a small, fast model or a large one that reasons for longer, and the difference shows on hard tasks. Your plan decides which models you can use at all, and on paid plans the heavier models use up your limits faster. Choosing on purpose gets you better answers and makes your allowance last.
The names help once you know the pattern. Within each family, the tiers run from small and fast to large and careful: Haiku, Sonnet and Opus in Claude, with Fable above them; Flash-Lite, Flash and Pro in Gemini, with Gemini 4 Argon as the flagship; and Luna, Sol and Astra among the GPT-6 models in ChatGPT. A bigger model is not always better. It is slower, uses more of your limit and adds little to a task a small model already does well.
What you need first
- Know your plan. Free and paid plans give access to different models, listed below.
- Know the task's stakes. How much would it cost you if the answer were slightly wrong?
- Find the model menu in your app, at the top of the chat or beside the message box.
The models you can actually choose in November 2026
| App | Free plan | Paid plans | Top model |
|---|---|---|---|
| ChatGPT (OpenAI) | GPT-6 Luna | GPT-6 Sol in chat on Plus, Pro, Business and Enterprise; GPT-6.1 Sol and GPT-6.1 Terra in ChatGPT Work and Codex | GPT-6 Astra, which powers Pro reasoning |
| Claude (Anthropic) | Claude Sonnet and Haiku | Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 | Claude Fable 5.1: credits on Pro, up to half the weekly limit on Max and Ultra |
| Gemini (Google) | Auto, which sends most prompts to Flash-Lite | Flash on Google AI Plus; Gemini 4 Argon, Pro and Flash on Google AI Pro and Ultra | Gemini 4 Argon |
| Perplexity | Basic models, chosen automatically | Model choice on Pro and Max, including GPT-6.1 Sol, Claude Opus 5.5 and Gemini 4 Argon in Computer | Depends on the task and Effort Mode |
Two catches stand out. ChatGPT's newest GPT-6.1 models live in ChatGPT Work and Codex rather than the plain chat tab. And since 9 October 2026, Gemini users without a plan are mostly answered by Flash-Lite, the fastest and smallest model.
How to choose the right AI model in 6 steps
- Ask what a slightly wrong answer would cost. Low stakes and you will read it anyway (a summary of a long thread, a rewrite, sorting feedback)? Use a fast model such as Claude Sonnet 5.5 or Haiku 5.5, GPT-6 Luna or Gemini Flash.
- Use the main model for judgement work. Reviewing a contract, analysing a spreadsheet, writing for a client: use Claude Opus 5.5, GPT-6 Sol or Gemini 4 Argon. Anthropic says Opus 5.5 performs at the level of Fable 5.1 on most work, and its own tests found it costs 40% less than Opus 5 on typical workloads at default settings, which is why it is a sensible default.
- Save the top model for problems the main one fails. If you have watched Opus or Sol struggle, switch to Fable 5.1 or ChatGPT's Pro reasoning for that task, then switch back.
- Turn up effort instead of switching models. Gemini lets you choose low, medium or high effort for each model; higher effort improves thoroughness but uses more of your limit. Perplexity's Effort Mode in Computer has Light, Standard, High and Ultra, and picks the model for you unless you choose Custom.
- Match the model to the job type. OpenAI's prompting guide recommends reasoning models for complex, multi-step tasks and faster GPT models for speed and cost, with more explicit instructions. For tasks where the AI clicks around apps or reads screenshots, OpenAI says GPT-6.1 Terra, rolling out in ChatGPT Work and Codex since 26 October 2026, supports tasks including visual reasoning, computer use and writing.
- Test your own task on two models. Run one weekly task on a fast and a strong model side by side. Most people find the fast one is fine for more work than they expected. See how to compare AI answers.
A worked example: one week, three models
A consultant on Claude Pro sets herself a simple rule for a week.
- Monday inbox triage and meeting-note clean-up: Claude Haiku 5.5. Fast, and she reads everything anyway.
- Tuesday client proposal from three source documents: Claude Opus 5.5, because the proposal needs judgement across a lot of material.
- Thursday financial model that Opus kept getting wrong: Claude Fable 5.1 on usage credits, for that one task only.
She stops hitting her limit mid-afternoon, and the work that matters gets the strongest model. The same pattern works in ChatGPT (Luna, Sol, Pro reasoning) and Gemini (Flash, Argon at medium effort, Argon at high effort).
Common mistakes when choosing a model and how to fix them
- Never opening the model menu. Fix: check what your default is today. It may have changed in October.
- Using the top model for everything. It burns your limit on emails. Fix: default to the main model and step down for routine work.
- Blaming the app for a free-model answer. Fix: remember free ChatGPT uses GPT-6 Luna and free Gemini mostly Flash-Lite.
- Choosing by benchmark charts. Labs publish their own numbers. Fix: judge on your own tasks. See what a benchmark is.
- Ignoring effort settings. Fix: try high effort on the same model before paying for a bigger one.
Which tool gives you the best model choice
- Claude gives the clearest choice: three tiers in one family, plus Fable. Read how to use Claude.
- ChatGPT is strongest if you want reasoning and agent models in one place, but the best ones sit in Work and Codex. Read how to use ChatGPT.
- Gemini has the most flexible effort control, and Google AI Pro adds a 1 million token context window. Read how to use Gemini.
- Perplexity lets paid users pick models from several labs. Read how to use Perplexity.
Not sure what a reasoning model is? See what is a reasoning model and what is an LLM. For a head-to-head, see ChatGPT vs Claude.
3 prompts to help you pick a model
Prompts to copy
Sort my tasks by model
Here are the AI tasks I do in a typical week: [list]. I use [app and plan]. Sort them into three groups: fast model is fine, main model needed, top model or high effort needed. Give a one-line reason for each and tell me which group probably uses most of my limit.
You get: A personal model plan for your week
Is this answer good enough?
You answered with a [fast] model. Here is the task and your answer: [paste]. List what a stronger model or higher effort would most likely improve, and whether those improvements matter for a [reader] who will [use it for]. Be honest if the answer is already good enough.
You get: A quick judgement on whether to rerun on a bigger model
Side-by-side model test
I will run this task on two models and paste both answers. Task: [task]. Answer A ([model]): [paste]. Answer B ([model]): [paste]. Compare them on accuracy, depth and usefulness, and tell me whether the difference justifies the heavier model for this kind of task.
You get: Evidence for your default model choice