Courseiva
1Z0-1127Chapter 6 of 18Objective 3.2

OCI Generative AI Models and Capabilities

OCI Generative AI models are a family of powerful tools that can generate human-like text, summarise long documents, understand the meaning of words, hold conversations, and even take actions on their own. For someone studying the 1Z0-1127 exam, understanding these distinct capabilities is crucial because the exam tests your ability to choose the right model for a specific business problem, not just a general understanding of AI.

12 min read
Intermediate
Updated Jul 23, 2026
Reviewed by Johnson Ajibi· Senior Network & Security Engineer · MSc IT Security

A simple way to picture OCI Generative AI Models and Capabilities

The Personal Chef Training School Analogy

Because you want to prepare a seven-course meal for a party, you first need to master different cooking skills. You wouldn't use a single recipe to make soup, a salad, and a dessert, because each dish requires a different technique. OCI Generative AI offers a suite of specialised models, each built for a distinct purpose, just like a chef learns separate skills for chopping, grilling, and baking.

You start by giving the chef a list of ingredients (your prompt). For a written menu (Text Generation), the chef writes down the entire description of each dish, from appetiser to dessert, in fluent language. For a shorter menu description (Summarisation), the chef reads your long recipe book and writes a single-paragraph summary that captures the essence. For finding similar menus (Embeddings), the chef describes the flavour profile of a dish as a unique set of numbers, then searches through thousands of other menus to find the ones with matching numbers, enabling you to recommend similar dishes to guests. For a quick back-and-forth (Chat), the chef answers your questions conversationally, remembering what you ordered last time. Finally, with an advanced kitchen team (Agent capabilities), the chef doesn't just describe the meal but also checks the pantry, orders missing ingredients, and adjusts the cooking time automatically. Each skill is distinct, yet they work together under the same kitchen roof, just as OCI models are separate but accessible through a single platform.

How It Actually Works

Generative AI refers to a type of artificial intelligence that creates new content instead of just analysing existing information. OCI (Oracle Cloud Infrastructure) Generative AI is a cloud service that gives you access to this capability. It is built around large language models, or LLMs, which are massive computer programs trained on billions of words from books, articles, and websites. Think of an LLM as a very educated assistant who has read almost everything ever written in English (or other languages) and can use that knowledge to predict the next most appropriate word in a sequence.

OCI offers several distinct types of these models, each tailored for a specific job:

Text Generation: This is the most famous capability. You give the model a starting prompt, like "Write an email to a customer apologising for a late delivery," and it generates the full text. It works by predicting one word at a time, always choosing the word that statistically fits best based on its training. This is useful for drafting emails, writing reports, creating marketing copy, or even writing code.

Summarisation: When you have a huge document, like a 100-page legal contract or a lengthy news article, the summarisation model can read it and produce a concise version, often just a few sentences or a short paragraph. It does this by identifying the most important sentences and key concepts, then rewriting them into a coherent summary. This saves enormous amounts of human reading time.

Embeddings: This is a more technical capability, but critical for many applications. An embedding is a mathematical representation of text in the form of a set of numbers, specifically a vector (a list of coordinates in a high-dimensional space). The key idea is that text with similar meaning gets placed close together in this mathematical space. For example, the sentence "I love my pet dog" and "My canine companion brings me joy" will have similar embedding vectors, even though they use different words. This allows you to perform semantic search (search by meaning) instead of just keyword search. If a customer asks a support chatbot "I need a refund," the embedding model can find the most relevant help article, even if the article never uses the word 'refund' and instead says 'return' or 'reimbursement'.

Chat: This is a conversational interface built on top of a text generation model, but optimised for back-and-forth dialogue. Unlike a one-shot text generation where you give one prompt and get one answer, a chat model remembers the history of the conversation. It keeps a running context of everything you and it have said in the current session. This enables natural follow-up questions like "Explain that again, but simpler" because the model remembers what 'that' refers to.

Agent Capabilities: This is the most advanced feature. An agent is an LLM that has been given the ability to use external tools and take actions. For example, you can give an agent access to a database of customer orders, an email system, and a calendar. You can then tell the agent, "Find all orders from last month that are still pending, email the customers, and schedule a call for next Tuesday." The agent will plan a sequence of steps, call the database tool to get the order list, use the email tool to send the messages, and use the calendar tool to book the appointments, all without human intervention for each step. This is what makes generative AI truly 'agentic' or capable of acting autonomously.

Each of these models exists because a single 'one-size-fits-all' model is inefficient and less effective for specialised tasks. By providing distinct model types, OCI allows you to choose the right tool for your specific use case, which saves money, reduces latency (delay), and improves accuracy.

Decision flowchart showing how a business problem leads to selecting one of the five OCI Generative AI model types, each with its own endpoint and output.

Walk-Through

1

Identify the Business Problem

Before touching any AI, you must define what you need: do you need to create new text, shorten existing text, find similar documents, hold a conversation, or automate a multi-step process? This step determines which model type to use.

2

Select the Model Type in OCI

In the OCI Console or via the API, you choose from dedicated endpoints for text generation, summarisation, embeddings, chat, or agent. Each endpoint is pre-configured for its specific task, so you don't need to train a model from scratch.

3

Prepare Your Input Data

You must format your input correctly. For text generation, you write a clear prompt. For embeddings, you pass a string of text. For agents, you define the tools (like a database connection) that the agent can call. Garbage in, garbage out.

4

Configure Parameters

You set parameters like maximum output length (in tokens), temperature (controls randomness of output), and top_p (another randomness control). For embeddings, you choose the embedding dimension. For agents, you define the tool calling policy.

5

Run Inference and Interpret Results

You call the model's endpoint (via API call or OCI Console). The model returns a response. For text generation, you get new text. For embeddings, you get a vector. For an agent, you may get a success message or a log of actions taken. You then integrate this response into your application.

6

Evaluate and Iterate

You check the quality of the output. Is the summary accurate? Does the chatbot answer correctly? Are the embedded search results relevant? Based on this, you may adjust the prompt, change parameters, or switch to a different model type (e.g., from basic text generation to a fine-tuned version).

What This Looks Like on the Job

An IT professional for a large online retailer needs to build a system that helps customer service agents respond to emails faster. Instead of manually reading each customer complaint, which can take minutes per email, they will use OCI Generative AI to automate parts of the process. - First, the IT professional uses the Embeddings model to create a smart search index. They take every past help article and customer response from the last five years and run each one through the embeddings model. This converts every article and response into a unique vector in the mathematical space. Now, when a new email complaint arrives, the system can instantly find the most relevant past responses by comparing the new email's embedding vector against the indexed vectors. - Second, they deploy a Summarisation model. When a customer email is a long, rambling story about a missing package, the summarisation model automatically generates a one-line summary like "Customer claims package marked delivered but not received, sent to wrong address." This summary is displayed to the human agent alongside the full email, giving them instant context. - Third, they build a Text Generation model to draft the reply. After the agent selects the appropriate help article found by the embeddings search, the system automatically generates a draft email that incorporates the customer's specific details, such as order number and name, using the text generation model. The human agent then just reviews and clicks send, saving 70% of their time. - Fourth, they add a Chat model to handle real-time queries. If a customer starts a live chat on the website, the chat model handles the initial greeting, asks for the order number, and even answers simple questions like "Where is my order?" It only escalates to a human when the conversation becomes complex. - Finally, they experiment with Agent capabilities. They build an agent that has access to the refund system. When a customer is clearly eligible for a refund (e.g., the item never arrived), the agent checks the refund policy, validates the customer's identity, and directly initiates the refund process in the backend system, then sends a confirmation email to the customer. This completely eliminates human involvement for straightforward cases. The IT professional's job here is to configure each model with the right parameters, such as the summarisation length or the maximum number of words in a generated text, and to ensure the different models safely connect to the company's internal databases and email systems.

How 1Z0-1127 Actually Tests This

The 1Z0-1127 exam directly tests your ability to distinguish between the different OCI Generative AI model types. You will see scenario-based questions where you are given a business requirement, and you must select the correct model. For example, a question might say 'A company wants to automatically shorten a 50-page annual report into a 500-word executive summary. Which OCI Generative AI capability should they use?' The correct answer is summarisation. - Key exam concepts to memorise:

Text Generation is for creating new text from scratch, like composing emails or code.

Summarisation is for condensing existing text into a shorter version.

Embeddings is for representing text as numbers to enable semantic search or clustering similar items.

Chat is for multi-turn dialogue with memory of the conversation history.

Agent capabilities are for tasks requiring the model to use external tools and perform actions autonomously.

Common traps the exam sets:

They will try to confuse Embeddings with Text Generation. If the problem involves 'finding similar documents' or 'searching by meaning', it is embeddings, not text generation.

They will try to confuse Chat with Text Generation. Chat always implies a dialogue history; if the question says 'single prompt, single response', it is text generation.

They will test that Agent capabilities are distinct from simple model inference. An agent must be able to call external systems. If the question only mentions generating a response, it is not an agent.

They often ask about the 'latency' or 'cost' implications. Embeddings are generally very fast and cheap per request compared to a full text generation call.

They may present a scenario where using an agent is overkill. For example, 'just writing a standard reply' does not require an agent; text generation or chat is sufficient.

Definitions you must memorise exactly:

Embedding: A vector representation of data that captures semantic meaning.

Inference: The process of the model generating a response.

Token: The basic unit of text an LLM reads, roughly a word or part of a word.

Context window: The amount of text the model can remember at once.

Fine-tuning: Training a pre-existing model on your own data to adapt it for a specific task. This is a separate concept but often tested alongside model types.

The exam will also ask about the benefits. For instance, using embeddings for a recommendation engine reduces the need for manual keyword tagging and works better for fuzzy matching. Agents reduce human workload by automating multi-step processes. Memorise these 'reason why' points.

Key Takeaways

OCI Generative AI includes five distinct model types: text generation, summarisation, embeddings, chat, and agent capabilities.

Text generation creates new content from a prompt; summarisation condenses existing content into a shorter form.

Embeddings convert text into numerical vectors for semantic search and similarity comparison.

Chat models maintain a conversation history, enabling coherent multi-turn dialogue.

Agent capabilities empower the model to use external tools and perform actions autonomously, like sending emails or querying databases.

Choosing the wrong model type for a task leads to poor performance and wasted compute resources.

The 1Z0-1127 exam tests scenario-based recognition of which model type fits a given business problem.

Fine-tuning a model is different from using pre-built model types; fine-tuning adapts a general model to a specific domain with your own data.

Easy to Mix Up

These come up on the exam all the time. Here's how to tell them apart.

Text Generation

Creates new text from a prompt.

Output is typically longer or equal to the input prompt.

Used for drafting, writing, or creative tasks.

Summarisation

Condenses existing text into a shorter version.

Output is always shorter than the input document.

Used for digesting long reports, articles, or transcripts.

Embeddings

Converts text to a numerical vector.

Does not generate conversational text.

Used for search, recommendation, and clustering.

Chat

Generates human-like text in a dialogue.

Maintains conversation history within a session.

Used for customer support, virtual assistants.

Chat Model

Only generates text responses.

Cannot call external tools or take actions.

Purely conversational, no task automation.

Agent Capabilities

Generates text AND can invoke external tools.

Can perform actions like sending emails or updating databases.

Used for autonomous task completion.

Standard Text Generation

Uses pre-trained base model without extra training.

Suitable for general-purpose tasks.

May not perform well on specialised jargon or domain-specific tasks.

Fine-tuned Text Generation

Trained further on specific company data.

Excels at tasks within that domain (e.g., medical, legal).

Requires additional data preparation and training cost.

Watch Out for These

Mistake

All OCI Generative AI models are the same; you just use one model for everything.

Correct

OCI offers distinct model types for different tasks: text generation, summarisation, embeddings, chat, and agents. Selecting the wrong type for a task leads to poor results or higher costs.

Beginners often hear 'AI model' as a single concept, so they assume one model can do everything. They don't realise that a summarisation model is optimised for condensing text, while a chat model is optimised for conversation memory.

Mistake

Embeddings are used to generate new text like a language model.

Correct

Embeddings are used to understand text by converting it into numbers, not to generate new text. They enable search and classification, not content creation.

The term 'embedding' sounds abstract and technical, so beginners often confuse it with 'embedding' words into a sentence, meaning generation. The mathematical nature is not intuitive.

Mistake

A Chat model is the same as a Text Generation model with a different name.

Correct

A Chat model is a specialised version that maintains conversation history. Text Generation models typically treat each prompt independently, without remembering previous exchanges.

Both can generate text, so it is easy to think they are identical. The distinguishing factor (memory of past turns) is a subtle but critical difference that beginners overlook.

Mistake

Agent capabilities are just a marketing term; they don't actually let the AI take actions on its own.

Correct

Agent capabilities are real and allow the LLM to call external tools (APIs, databases) and perform actions autonomously, such as sending emails or updating records, without human intervention at each step.

Many people have only seen chatbots that give answers but don't do anything, so they assume 'agent' is hype. They underestimate how much orchestration and tool-calling is happening under the hood.

Mistake

The Summarisation model will perfectly capture every detail from the original document.

Correct

The summarisation model generates a condensed version and may omit details or lose nuance. It is designed to preserve the main points, not every fact.

Beginners expect AI to be perfect and are disappointed when a summary misses a minor but important clause. They don't understand the trade-off between brevity and completeness.

Do You Actually Know This?

Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.

Frequently Asked Questions

What is the difference between text generation and summarisation models?

Text generation creates entirely new text from a prompt (like writing an email), while summarisation condenses an existing long document into a shorter version. You use text generation for creation and summarisation for reduction.

Do I need to train my own model to use OCI Generative AI embeddings?

No. OCI provides pre-built embedding models that you can use directly. You just send text to the endpoint, and it returns the embedding vector. Training (fine-tuning) is a separate, optional step for specialised needs.

Can a chat model remember what I said in a previous session?

Not automatically. Chat models remember the conversation within a single session, but they do not remember across sessions unless you save the conversation history and feed it back into the model as context each time.

How does an agent 'take action' if it's just a language model?

An agent is given access to external tools (like an API to send emails). When you give it a task, it generates a plan in the form of tool calls, then the system executes those calls. The model itself doesn't execute code; it decides which tool to use and with what parameters.

What is a token and why does it matter for these models?

A token is a piece of text, roughly a word or part of a word, that the model reads. Models have a maximum token limit (context window). If your prompt or conversation exceeds this limit, the model will forget earlier parts. You must manage input length to stay within limits.

Is it expensive to run these models on OCI?

Cost depends on usage. Embeddings are generally cheaper and faster per request because they use smaller, more efficient models. Text generation and summarisation cost more per token of output. Agent capabilities can incur additional costs for external tool calls. You pay for compute and token usage, not a flat fee.

Terms Worth Knowing

Keep going

You've finished OCI Generative AI Models and Capabilities. Continue through the 1Z0-1127 study guide to build a complete picture of the exam.

Done with this chapter?