The Model Landscape
Open, closed, big and small.
There are now hundreds of models, from many companies. How do you choose? In this lesson, you will learn the main differences, closed versus open, large versus small, and a simple way to match a model to your task.
- Hundreds of models to choose from
- Closed or open?
- A simple way to choose
Closed versus open models
Closed models, like the ones behind many popular chat apps, are used through a service. You send a request, and get an answer. They are usually the most capable, and easy to start with, but you depend on the provider, and your data leaves your computer. Open weight models, like Llama and Mistral, can be downloaded. You can run them on your own hardware, adapt them freely, and keep your data private, but you need the technical skill, and the computing power.
Closed models
- Use through an online service
- Often the most capable
- Easy to start with
- Data goes to the provider
Open-weight models
- Download and run yourself
- Adapt freely, fine-tune
- Data can stay private
- Needs hardware and skill
Big models versus small models
Size matters too. Large models know more, and reason better, but they are slower, and cost more to run. Small models are fast, cheap, and can even run on a laptop, or a phone. They may be less capable on hard tasks, but for simple jobs, like classifying emails, or extracting names, they are often plenty. A smart team uses a large model for hard problems, and a small one for the rest.
Large models
- Know more, reason better
- Slower and more expensive
- Best for hard problems
Small models
- Fast and cheap
- Can run on a laptop or phone
- Great for simple, repeated tasks
A simple way to choose
Here is a simple way to choose. Ask four questions. How hard is the task? Hard tasks need a stronger model. How private is the data? Sensitive data may need a model you run yourself. How fast must it respond? Speed favours small models. And how much can you spend? Costs add up quickly at scale. Always test two or three models, on your own real examples, because rankings online may not match your needs.
| Question | If the answer is yes |
|---|---|
| Is the task hard? | choose a larger, stronger model |
| Is the data sensitive? | consider an open model you host |
| Must it be fast? | choose a smaller model |
| Is the budget tight? | start small, scale only if needed |
Summary
Excellent. Closed models are capable and convenient, open models give you control and privacy. Large models are stronger, and small models are fast and cheap. Choose by difficulty, privacy, speed, and cost, and always test on your own examples. Next, the newest idea, models that think before they answer.
Key takeaways- Closed = convenient, open = control
- Large = strong, small = fast and cheap
- Test on your own examples