---
id: con-model-choice
schema: concept/1
layer: garden
status: established
name: Model choice
aka: ["model picker", "choosing a model", "which AI should I use", "model selector"]
summary: >
  Picking which AI model, or which tool, to use for a given job, based on what it is good at, how
  fast and costly it is, and how much is at stake.
first_seen: "2023"
relations:
  builds_on: [con-large-language-model]
  evolved_from: []
  related: [con-reasoning-models, con-jagged-frontier, con-evals]
  parallels:
    - idea: "matching who you assign to how hard the task is"
      field: "management"
      note: "You don't send your most senior analyst to reformat a spreadsheet, or a new hire to negotiate a contract."
    - idea: "getting a second opinion"
      field: "medicine"
implementations:
  - { provider: openai, name: "ChatGPT model picker" }
  - { provider: anthropic, name: "Claude model selector" }
  - { provider: perplexity, name: "Perplexity model selection" }
  - { provider: huggingface, name: "HuggingChat model list", resource: res-huggingface-huggingchat }
competencies: [cmp-frame-the-outcome, cmp-check-the-result]
origin:
  kind: ai-proposed
  by: "seed agent 2026-10-16"
  rationale: "Learners often blame AI in general when a different model or tool would have done the job."
---

Most AI tools let you pick between models: a fast everyday one, a slower one that thinks harder, sometimes a cheaper or older one. Different companies' tools also have different strengths. Model choice is the habit of matching the tool to the job instead of using whatever opens by default.

A quick, light model is fine for rewording a paragraph. A careful reasoning model is worth the wait for a tricky analysis. For anything that matters, trying the same task in two tools and comparing is one of the fastest ways to learn what each one does well and where it slips.

You don't need to track every release. A small personal test, run on a real task you do often, tells you more than any leaderboard.

## Example

A grants officer pastes the same funder guidelines into two different AI tools and asks each for a checklist of eligibility rules. One misses a deadline buried in a footnote. She now uses the other for compliance reading and keeps the first for brainstorming.

## How it connects

- **Reasoning models** are one option you choose when a task needs careful thought.
- The **jagged frontier** explains why the best tool varies task by task.
- A tiny personal test set is the start of **evals**.

## Try it

Pick one task you did this week and guess which of two tools will handle it better, and why. Run it in both and see if you were right.