01 // Resource guide
Updated June 2026AI models
Understand what a model is, what changes between model tiers, and how to choose enough capability for the job without wasting time or money.
Bottom line
Match the model to the job, not the hype. Start with the default, then change only when speed, cost, privacy, or quality gives you a reason.
01 // City map
The model is the building
A company owns a plot in the AI city. It builds a family of models on that plot. The app or harness you use is the storefront where you enter and work.
Company
01The organization building and operating the model family.
Model
02The underlying intelligence that interprets inputs and generates outputs.
Harness
03The chat, agent, coding tool, or app that gives you access to a model.
02 // Model jobs
Most choices fall into four lanes
Capability-first
01Use for difficult reasoning, ambiguous planning, complex synthesis, or work where a weak answer creates real cost.
Balanced default
02The best starting point for normal writing, analysis, file work, and everyday agent tasks.
Fast and efficient
03Use for simple transformations, classification, extraction, and high-volume repeat work.
Specialized or local
04Use when the job needs a specific medium, tool, privacy boundary, or self-hosted model.
03 // Five dials
Compare trade-offs that survive model-name changes
| Dial | Question to ask | When it matters |
|---|---|---|
| Capability | Can it reliably handle this level of reasoning and ambiguity? | Complex or high-consequence work |
| Speed | How quickly does it respond or finish an agent task? | Interactive and repeated work |
| Cost / usage | Is the quality gain worth the extra usage? | Large files, long sessions, automation |
| Context | Can it use all the files and instructions needed at once? | Research, codebases, long histories |
| Privacy / control | Where is data processed and what can the system retain or access? | Personal, school, or restricted information |
04 // Quick choice
A three-step model decision
- 01
Start with the recommended default
The provider usually chooses a balanced model for normal use. Do not optimize before you have a problem.
- 02
Run one representative task
Use a real example, a clear definition of done, and the same prompt when comparing alternatives.
- 03
Change one dial
Move up for quality, down for speed or usage, or local for control. Keep the rest of the workflow unchanged.
- Do not use the largest model for every small task.
- Do not treat benchmark rank as proof that a model fits your workflow.
- Do not confuse the model with the app, plan, connector, or agent using it.
Official sources