Fundamentals of gen AI · Topic 3 of 6
Choosing a Foundation Model and Gen AI Use Cases
How to choose a foundation model by modality, context window, security, reliability, cost, performance and customization, and where gen AI creates business value.
- 4 min
- Factors: modality, context window, security, availability and reliability, cost, performance, customization.
- Context window = how much input and output the model handles at once.
- Match model size to task complexity and volume.
What it is
There is no single best model. Choosing a foundation model means matching its abilities to the business use case and its constraints. The exam lists the factors: modality, context window, security, availability and reliability, cost, performance, and fine-tuning and customization. It also expects you to recognize where gen AI adds value: to create, summarize, discover and automate.
How it works
The selection factors
| Factor | Question to ask | Example |
|---|---|---|
| Modality | What types of input and output are needed: text, image, audio, video, code? | Reading scanned invoices needs a model that understands images |
| Context window | How much information must the model consider at once? | Summarizing a 300-page contract needs a long context window |
| Security | Where does the data go, who can access it, does it meet our policies? | Sensitive customer data must stay under enterprise controls |
| Availability and reliability | Is the model available in our regions, with the uptime we need? | A customer-facing assistant needs a dependable service |
| Cost | What will it cost per request at our volume? | High-volume chat may favor a smaller, cheaper model |
| Performance | Is it accurate and fast enough for the task? | Live chat needs low latency; legal review needs high quality |
| Fine-tuning and customization | Can we adapt it to our data and style? | Matching a brand's tone or a domain's vocabulary |
The context window is the amount of text (measured in tokens) the model can take into account in one request, including the prompt and the output.
Bigger is not always better. A smaller, faster model often wins when the task is simple and volume is high. Reserve the largest models for complex reasoning.
Tradeoffs in practice
| Situation | Usually favors |
|---|---|
| Simple, high-volume tasks such as classification or short replies | Smaller, faster, cheaper models |
| Complex reasoning over long documents | Larger, more capable models with long context |
| Strict data control or offline use | Models that can be deployed in the organization's own environment, such as open models |
| Images, audio or video involved | Multimodal models |
What gen AI can do for a business
| Capability | What it means | Examples |
|---|---|---|
| Create | Generate new content | Text generation (marketing copy, emails), image generation (product visuals), code generation (developer assistance), video generation (training clips) |
| Summarize | Condense large amounts of information | Meeting notes, call transcripts, long reports, customer reviews |
| Discover | Find and connect information | Search across company documents, answer questions grounded in internal knowledge, data analysis in plain language |
| Automate | Complete tasks and workflows | Agents that process claims, draft replies, update records; personalized user experiences at scale |
Matching use cases to value
Strong first use cases share three traits: clear business value, available data, and acceptable risk. Internal productivity uses (summarizing documents for staff) often start first because mistakes are caught by employees before they reach customers.
A worked scenario
A law firm wants to summarize long case files, draft client letters and search its archive. Case files run to hundreds of pages, so it needs a long context window. Client data is confidential, so security and enterprise controls matter most. Drafting letters in the firm's style benefits from customization. Volume is moderate, so a capable model is affordable. The use cases map to summarize, create and discover.
How the exam asks about this
- A requirement is given (long documents, images, low cost, strict data control) and you choose the deciding factor.
- A business task is given and you classify it as create, summarize, discover or automate.
- "Which model for a simple, high-volume task?" A smaller, faster, lower-cost model.
What to remember for the exam
- Factors: modality, context window, security, availability and reliability, cost, performance, customization.
- Context window = how much input and output the model handles at once.
- Match model size to task complexity and volume.
- Gen AI value: create, summarize, discover, automate.
Do not choose a model on benchmark quality alone. Cost, latency, security and reliability decide whether it works in production.
Common mistakes
Always choosing the largest model. It can be slower and costlier than the task needs.
Ignoring modality. A text-only model cannot read a scanned image.
Forgetting security and data location. These can rule out a model regardless of quality.
Check yourself
Which factor matters most when summarizing very long documents?
The context window.
Classify "answer employee questions using the company's internal documents".
Discover.
Why might a business choose a smaller model?
For simple, high-volume tasks it is faster and cheaper while still good enough.
Topic 3 of 6 in Fundamentals of gen AI
Test yourself on Fundamentals of gen AI
The full Google Cloud Generative AI Leader question bank has 512 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Google Cloud. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.