Overview
Domain 6 shows how Oracle turns generative AI concepts into products. The first part is the OCI Generative AI service: a fully managed service offering pretrained chat models from providers such as Cohere and Meta, plus embedding models, available on demand or on dedicated AI clusters that also host fine-tuned custom models. The second part is Select AI in Autonomous AI Database, which lets users ask questions in plain language and have an LLM generate, run or explain SQL against their schemas.
The third part is AI Vector Search, which stores embeddings in a native VECTOR column type and finds similar content with distance functions and vector indexes. Oracle renamed Oracle Database 23ai to Oracle AI Database 26ai in late 2025; the section title on secondary sources still uses the older name, but the features are the same family.
How it's tested
Expect questions that name an Oracle feature and ask what it does, or describe a goal and ask which feature, setting or action achieves it. Choices often differ by one product detail, such as on-demand versus dedicated hosting, or showsql versus runsql, so precise recall matters more than broad theory.
-
Hosting choice
You choose on-demand inference or a dedicated AI cluster based on fine-tuning needs, isolation and predictable throughput.
-
Select AI action
A user wants to run a query, see the SQL, have it explained or get a narrated answer, and you choose the matching action.
-
Vector search mechanics
A similarity search or RAG scenario tests embeddings, distance metrics and approximate versus exact search.
A finance team wants to check what SQL Select AI would generate for a question before anything runs against production tables. The runsql action executes the query, and narrate returns a prose answer, so neither meets the requirement. The showsql action returns the generated SQL statement without executing it, which lets the team review it first. That is the correct answer.
Objectives 3
Official objectives from the Oracle outline, in outline order.
6.1 Describe OCI Generative AI servicesManaged LLMs, embeddings, hosting modes and fine-tuning.
- What it means
- OCI Generative AI is a fully managed service that provides access to large language models through a console playground, APIs and SDKs. It offers chat models for generation and conversation, and embedding models that convert text into vectors for semantic search.
- What to understand
- Pretrained models come from providers including Cohere and Meta, with other model families added over time. On-demand mode is pay-as-you-go on shared capacity. Dedicated AI clusters reserve GPU capacity for a tenancy, are required for fine-tuning, and host custom models behind endpoints. Fine-tuning uses parameter-efficient methods such as T-Few and LoRA. Customer data and prompts are not shared with model providers, and generation settings include temperature, top p, top k and maximum output tokens.
- How it shows up
- Expect questions on when a dedicated AI cluster is needed and what the embedding models are for.
- Common confusion
- Thinking fine-tuning runs in on-demand mode; it needs a dedicated AI cluster.
6.2 Discuss Autonomous Database Select AINatural language to SQL and chat inside Autonomous AI Database.
- What it means
- Select AI lets users write a question in natural language after the keywords SELECT AI. An LLM uses schema metadata to generate SQL, which the database can run, display or explain. An AI profile defines the provider, credentials, model and which objects the LLM may see.
- What to understand
- Profiles are created with the DBMS_CLOUD_AI package and set for a session. Supported providers include OCI Generative AI and several third-party services. Key actions are runsql, the default, which executes the generated query; showsql, which only returns the SQL; explainsql, which describes the query in plain language; narrate, which returns the result as prose; and chat, which sends the prompt straight to the LLM. Select AI can also use a vector index for retrieval-augmented answers over documents.
- How it shows up
- A requirement such as reviewing SQL before execution is matched to the correct Select AI action.
- Common confusion
- Assuming Select AI sends table rows to the LLM to write SQL; generation relies on metadata.
6.3 Discuss Oracle Vector SearchEmbeddings, VECTOR columns, distance functions and indexes.
- What it means
- Oracle AI Vector Search stores embeddings in a native VECTOR data type alongside relational data and finds semantically similar items by measuring distance between vectors. It is the retrieval layer for RAG built in the database.
- What to understand
- Embeddings can be generated in the database from an imported ONNX model or by an external service such as OCI Generative AI. The VECTOR_DISTANCE function supports metrics including cosine, Euclidean and dot product, and the metric should match how the model was trained. Exact search compares every vector; approximate search uses vector indexes, such as in-memory neighbor graph (HNSW) or neighbor partition (IVF) indexes, trading a little accuracy for speed. Query and stored vectors must come from the same model.
- How it shows up
- Questions test what an embedding represents, the purpose of vector indexes, or how RAG uses vector search.
- Common confusion
- Mixing embeddings from two different models and expecting meaningful distances.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
OCI Generative AI
Managed service offering chat and embedding LLMs.
- Remember
- Playground, API and SDK access.
- Tested as
- Service identification.
- Trap
- Confusing it with OCI Data Science.
-
Chat models
Generate text and hold conversations.
- Remember
- Used for drafting, Q&A and summarising.
- Tested as
- Model type questions.
- Trap
- Using them to create search vectors.
-
Embedding models
Turn text into numeric vectors capturing meaning.
- Remember
- Feed semantic search and RAG.
- Tested as
- Model type questions.
- Trap
- Expecting them to produce readable text.
-
On-demand mode
Pay-per-use inference on shared capacity.
- Remember
- No cluster to manage.
- Tested as
- Hosting choice items.
- Trap
- Assuming it supports fine-tuning.
-
Dedicated AI cluster
GPU capacity reserved for one tenancy.
- Remember
- Needed for fine-tuning and hosting custom models.
- Tested as
- Hosting choice items.
- Trap
- Thinking it is required for every request.
-
AI profile
Select AI configuration for provider, model, credential and allowed objects.
- Remember
- Created with DBMS_CLOUD_AI.
- Tested as
- Select AI setup questions.
- Trap
- Forgetting the profile limits which tables are exposed.
-
runsql
Default Select AI action that executes generated SQL.
- Remember
- Returns query results.
- Tested as
- Action-matching items.
- Trap
- Assuming the default only displays SQL.
-
showsql
Returns generated SQL without running it.
- Remember
- Use for review.
- Tested as
- Action-matching items.
- Trap
- Confusing it with explainsql.
-
narrate
Returns query results as a natural language answer.
- Remember
- Results are summarised by the LLM.
- Tested as
- Action-matching items.
- Trap
- Expecting raw rows.
-
VECTOR data type
Native column type for embeddings.
- Remember
- Vectors live next to business data.
- Tested as
- Vector Search basics.
- Trap
- Assuming a separate vector database is required.
-
Distance metric
Cosine, Euclidean or dot product measure similarity.
- Remember
- Match the metric to the embedding model.
- Tested as
- Search mechanics.
- Trap
- Thinking metric choice never affects results.
-
Vector index
HNSW or IVF structures for approximate search.
- Remember
- Faster, slightly less exact.
- Tested as
- Performance questions.
- Trap
- Assuming approximate search returns identical results to exact search.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
OCI Generative AI hosting
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| On-demand | Shared capacity, pay per use. | No reservation, pretrained models only. | Trying to fine-tune here. |
| Dedicated AI cluster | Reserved GPUs for one tenancy. | Fine-tuning, custom models, predictable throughput. | Choosing it for light, occasional use. |
Select AI actions
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| runsql | Generate and execute SQL. | Returns rows. | Using it when SQL must be reviewed first. |
| showsql | Generate SQL only. | Nothing runs. | Thinking it explains the logic. |
| explainsql | Describe the generated SQL in plain language. | Focus on understanding the query. | Expecting results. |
| narrate | Run and summarise results in prose. | Answer reads as text. | Expecting a table. |
| chat | Send the prompt straight to the LLM. | No SQL generation. | Expecting it to query your tables. |
Search styles
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Keyword search | Matches exact terms. | Misses paraphrases. | Expecting it to find synonyms. |
| Exact vector search | Compares the query with every vector. | Most precise, slower at scale. | Using it on huge tables with tight latency. |
| Approximate vector search | Uses a vector index. | Faster with slight accuracy trade-off. | Expecting perfect recall. |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Fine-tune on demand
- What candidates pick
- On-demand mode
- Why it looks right
- It looks simpler and cheaper.
- Why it's wrong
- Fine-tuning and hosting custom models need GPU capacity reserved through a dedicated AI cluster.
Clue to watch for Mentions of custom or fine-tuned models.
-
Show versus explain
- What candidates pick
- explainsql to see the query
- Why it looks right
- Explaining sounds like showing.
- Why it's wrong
- explainsql describes the query in natural language, while showsql returns the actual SQL text for review.
Clue to watch for Requirement to view or copy the SQL.
-
Data sent for SQL generation
- What candidates pick
- Select AI uploads table rows to the LLM
- Why it looks right
- The LLM seems to need the data.
- Why it's wrong
- SQL generation relies on schema metadata such as table and column names, not on reading table contents.
Clue to watch for Data privacy concern in the scenario.
-
Chat queries tables
- What candidates pick
- Use the chat action for database answers
- Why it looks right
- Chat sounds conversational and capable.
- Why it's wrong
- The chat action passes the prompt directly to the LLM without generating SQL against your schema.
Clue to watch for Answer must reflect current table data.
-
Mixed embedding models
- What candidates pick
- Reuse stored vectors after switching models
- Why it looks right
- Re-embedding looks expensive.
- Why it's wrong
- Different models produce incompatible vector spaces, so stored content must be re-embedded with the new model.
Clue to watch for A change of embedding model.
-
Approximate equals exact
- What candidates pick
- Vector index gives identical results
- Why it looks right
- Indexes usually just add speed.
- Why it's wrong
- Approximate indexes trade some accuracy for speed, so results can differ slightly from an exhaustive search.
Clue to watch for Target accuracy or latency requirement.
Answer framework
Apply these steps when a Domain 6 question asks which Oracle generative AI feature or setting to use.
- Decide whether the goal is text generation, semantic representation, natural language querying of tables, or similarity search over documents.
- For generation or embeddings from OCI Generative AI, pick on-demand for standard use and dedicated AI clusters for fine-tuning or custom models.
- For questions over relational data in Autonomous AI Database, choose Select AI and match the action to whether SQL should run, show, explain or narrate.
- For finding similar content or grounding an LLM, choose AI Vector Search with embeddings from one consistent model and a fitting distance metric.
- Check performance and accuracy needs to decide between exact search and an approximate vector index such as HNSW or IVF.
Practice question
Want more like this? The 1Z0-1122-26 practice test has questions for every domain.
Checklist
0 of 12
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Plan five to seven hours, more if you have never used Oracle's database features. Watch the OCI Generative AI and Oracle database AI lessons in the free course, then read the overview pages of the OCI Generative AI documentation and the Select AI documentation for Autonomous AI Database.
Hands-on time helps most here. In the Generative AI playground, try a chat model and an embedding model and note what each returns. If you can create an Autonomous AI Database, set up an AI profile and run the same question with showsql, runsql and narrate to see the differences. Oracle LiveLabs workshops cover both without needing much setup.
For Vector Search, write a short flow on paper: chunk documents, embed them, store them in a VECTOR column, embed the question with the same model, rank by distance, pass the top chunks to an LLM. Practise explaining each step.
Finish with practice items and keep a list of product facts you missed, since this section rewards exact recall.
FAQ
How much of the exam does OCI Generative AI and Oracle 23ai represent?
Oracle has not published weights for 1Z0-1122-26. Secondary sources estimate this section at about four of the 40 questions, but its product-specific items can be decisive.
Is it Oracle Database 23ai or Oracle AI Database 26ai?
Oracle renamed 23ai to Oracle AI Database 26ai in late 2025. Secondary sources keep the older section name; the AI features covered are the same family.
Do I need to write SQL for Select AI questions?
No. Know the purpose of the AI profile and what each action returns. Recognising the SELECT AI syntax is enough.
Which models does OCI Generative AI offer?
Chat and embedding models from providers including Cohere and Meta, with Oracle adding other families over time. Check the current model list before the exam.
Is retrieval-augmented generation tested here?
Yes, at a conceptual level: vector search retrieves relevant content and the LLM answers using it, which reduces hallucination.
Do I need hands-on experience?
It is not required, but a short LiveLabs workshop on Select AI or Vector Search makes the product facts much easier to remember.
Other Artificial Intelligence & Machine Learning certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 6 knowledge
You have reviewed OCI Generative AI, Select AI and AI Vector Search. Check your recall of the product details these items depend on.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the 1Z0-1122-26 AI tutor Get any concept explained your way