Which AI model for which task?
Which AI model for which task? Use the smallest model that does the task well. A fast model handles short and frequent jobs. A balanced model handles everyday work with several steps. A deep-thinking model is for hard problems, and a local model is for data that must stay in-house.
Why there is more than one model
An AI model is the program that does the thinking behind an AI app. Every AI company offers several, from small to large. The large ones can handle harder problems, but they answer more slowly and cost more each time you use them.
It is like choosing a vehicle. You do not hire a lorry to deliver a letter, and you do not move house on a bicycle. The right choice depends on the load.
The four kinds
- Fast. The smallest models. They answer almost at once and cost the least. They follow clear instructions well and struggle with tasks that need many steps of thought.
- Balanced. The middle of the range and a sensible default. They plan, write well and can look things up, at a moderate price and speed.
- Deep-thinking. The largest models. They work through a problem before they answer. They are the slowest and cost the most.
- Local. Models that run on your own computers. Nothing leaves your company.
Model names change every few months. These four kinds stay.
Eight tasks and the model that suits each
| Task | Model |
|---|---|
| Sort incoming requests by topic | Fast |
| Pull the date, amount and supplier from an invoice | Fast |
| Translate a short notice | Fast |
| Summarise a long email exchange with a customer | Balanced |
| Draft a reply in your company's tone | Balanced |
| Plan a project into phases and tasks | Balanced |
| Compare two long contracts point by point | Deep-thinking |
| Summarise personnel files that may not leave the company | Local |
How to choose, in four steps
- Collect ten real examples. Take them from last month's work and include two or three awkward ones.
- Start with the fast model. Run all ten and count how many results you would have accepted.
- Improve the instructions first. Unclear instructions cause more poor results than a model that is too small.
- Move up one step only if the problems remain. Then run the same ten examples again and compare.
Repeat the test about once a year. New models arrive often, and a task that needed a balanced model last year may run well on a fast one now.
What a bigger model does not fix
Two things shape a result more than the size of the model. The first is the instructions: one clear job and a clear idea of what the answer should look like. The second is the information the model is given. No model can know what it was not shown. If it has never seen your price list, even the largest one can only guess.
One model for the whole company?
It is tempting to pick the strongest model and use it for everything. That makes every small task slower and more costly than it needs to be. Choosing per task works better. The sorting job that runs hundreds of times a day gets the fast model. The contract comparison that runs twice a month gets the deep-thinking one.
How Worknano handles this
In Worknano the model is a setting on each agent, next to its instructions. An agent is an AI helper you set up for one job. The guide for building agents gives the same rule of thumb: a fast model suits short checks, and a balanced one suits planning with several steps. You are not tied to a single AI company. You can connect any model you choose, including local models, open-source models or your company's own private endpoint, and match the model to the task, your budget and your data rules.
- Which AI model does the AI module builder use?
- AI models and where your data is processed
- Build your own agent: model and instructions
Create a free workspace and set the model for your first agent.