Answer generation model versions and lifecycle
Stay organized with collections
Save and categorize content based on your preferences.
Vertex AI Search offers multiple model versions for you to choose when
generating answers. You can choose the model versions when using search
summaries and answers and
follow-ups.
Available models
Vertex AI Search uses two types of models for question and answering
use cases:
- Vertex AI LLM models that have been tested on question and
answering tasks
- Vertex AI Search models that are based on Vertex AI LLM
models and further trained to address question and answering tasks
Vertex AI Search models share the same discontinuation date as their
base Vertex AI LLM models. The base LLM model is available for six
months after the release date of the next version of the model, per the Vertex
AI model lifecycle
policy.
Leave enough time to migrate to new models before the discontinuation dates.
The following table lists model version specifications. When you set a model
specification, the API uses the specified model to generate answers.
| Model version |
Description |
Context window |
Discontinuation date |
stable |
The default model choice if the model version is not set.
The stable model specification points to
gemini-2.0-flash-001/answer_gen/v1.
The model designated as stable changes periodically as new
models and versions become available.
|
128K |
N/A |
gemini-2.0-flash-001/answer_gen/v1 |
A Vertex AI Search model based on the gemini-2.0-flash-001 model with additional tuning to address question and answering tasks.
The model is frozen after release.
|
128K |
Feb 5, 2026 |
gemini-1.5-flash-002/answer_gen/v1 |
A Vertex AI Search model based on the gemini-1.5-flash-002 model with additional tuning to address question and answering tasks.
The model is frozen after release.
|
128K |
Sept 24, 2025 |
gemini-1.5-flash-001/answer_gen/v2 |
A Vertex AI Search model based on the gemini-1.5-flash-001 model with additional tuning (version 2) on blended structured and unstructured data to address question and answering tasks.
The model is frozen after release.
|
128K |
May 24, 2025 |
gemini-1.5-flash-001/answer_gen/v1 |
A Vertex AI Search model based on the gemini-1.5-flash-001 model with additional tuning to address question and answering tasks.
The model is frozen after release.
|
128K |
May 24, 2025 |
preview |
The preview model specification points to the latest gemini-1.5-pro-002
model. The preview model is subject to change without notification. If
you use preview as the model, you might see changes in the responses when the
model changes. If you want consistency in the responses, select a
specific model.
|
128K |
N/A |
What's next
Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.
Last updated 2025-05-15 UTC.
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Hard to understand","hardToUnderstand","thumb-down"],["Incorrect information or sample code","incorrectInformationOrSampleCode","thumb-down"],["Missing the information/samples I need","missingTheInformationSamplesINeed","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2025-05-15 UTC."],[[["Agentspace Enterprise provides multiple model versions for generating answers, which can be selected when using search summaries and answers/follow-ups."],["The models available include Vertex AI LLM models and Agentspace Enterprise models that are trained for question and answering tasks."],["The `stable` model is the default choice, pointing to the latest `gemini-1.5-flash-001/answer_gen/v1` but this may change when newer versions are available."],["Each model version, except for the `stable` and `preview` models, has a specific discontinuation date, allowing users to migrate to newer models before that date."],["The `preview` model points to the most current `gemini-1.5-pro-002` and is subject to change without prior notification, potentially affecting response consistency."]]],[]]
