It has been revealed that the on-premises Enterprise Edition (Linux x86-64) of “Oracle AI Database 26ai,” which will be the next Long Term Support (LTS) release of the Oracle Database, is scheduled for release in January 2026.
This article provides a flash report from a practitioner’s perspective on the long-awaited on-premises release information, the decisive differences from the current mainstream Oracle Database 19c, and implementation examples of the highlight feature, AI Vector Search.
Article Summary (TL;DR)
- Conclusion: The on-premises version of Oracle 26ai (formerly 23ai) is scheduled to appear in late January 2026. It will be the major upgrade destination from 19c.
- Biggest Change: AI Vector Search is included as standard. RAG (Retrieval-Augmented Generation) can be implemented via SQL.
- Note: The CDB/PDB architecture is strictly mandatory. Since the Non-CDB configuration has been abolished, migration design is necessary.
- SQL Evolution: Developer Experience (DX) is significantly improved with features like eliminating
FROM DUALand support for theBOOLEANdata type.
1. What is Oracle AI Database 26ai?
Name Change from 23ai to 26ai
Although previously known as “Oracle Database 23ai,” an announcement in October 2025 unified the name of the upcoming LTS (Long Term Support) version to “Oracle AI Database 26ai”.
The internal version is 23.26.x, making it effectively an enhanced version of 23ai.
Release Schedule and Support Period
| Item | Details |
| Product Name | Oracle AI Database 26ai Enterprise Edition |
| Target OS | Oracle Linux 8 / 9, Red Hat Enterprise Linux 8 / 9 (x86-64 first) |
| Release Date | January 2026 (Planned) Likely around the CPU release date (Jan 20) |
| Support Period | Premier Support: Until Dec 31, 2031 (Planned) |
| Status of 19c | Premier Support has been extended until the end of 2029 |
Developer Free Version Available in Advance
Ahead of the official Enterprise Edition release, “Oracle Database 26ai Free” (equivalent to the former Express Edition) for learning, verification, and small-scale development is already available.
It can be obtained as a Docker container image or RPM package, allowing you to try out new features like AI Vector Search in your local environment immediately.
- Limits: Max 12GB user data, 2 CPU threads, 2GB RAM
- Usage: Development, Testing, Learning (Not for Production)
- Download: AI Database 26ai Free | Oracle
2. Differences Between 19c and 26ai (Comparison Table)
We have summarized the changes with the greatest impact when considering migration from 19c.
| Feature Category | Oracle 19c | Oracle 26ai (formerly 23ai) | Impact / Benefit |
| Architecture | Non-CDB / CDB Coexistence | CDB/PDB Configuration Only | Non-CDB is abolished. Configuration change is mandatory during migration. |
| AI & Search | Oracle Text (Full-text search) | AI Vector Search | Natively stores vector embeddings. RAG construction is possible within the DB alone. |
| Data Types | No BOOLEAN (PL/SQL only) | BOOLEAN Type Support | TRUE/FALSE can be defined in table columns. |
| SQL Syntax | FROM DUAL Mandatory | FROM DUAL Optional | Executable with just SELECT sysdate;. |
| Dev Features | JSON (BLOB/CLOB) | JSON Relational Duality | Allows the best of both JSON and relational tables. |
| Management | Password File Management | Simplified Management | PDB-level password files, etc., become more flexible. |
3. Practice: 26ai New Feature Implementation Examples for 19c Users
Here, we introduce features added in 26ai using SQL examples that are easy to understand for 19c users.
① Implementation of AI Vector Search
This is the highlight feature of 26ai. It enables “semantic search” within the Oracle Database without using an external vector DB.
Scenario: Vectorize product review text and search it.
-- 1. Create a table with a vector type column
-- VECTOR type: Specify dimensions (e.g., 3) and number format (FLOAT32)
CREATE TABLE product_reviews (
review_id NUMBER PRIMARY KEY,
review_text VARCHAR2(1000),
embedding VECTOR(3, FLOAT32)
);
-- 2. Insert data
-- Normally vector values generated by an LLM are inserted, but dummy values are set here
INSERT INTO product_reviews VALUES (
1,
'The battery life is excellent.',
'[0.1, 0.8, 0.2]'
);
INSERT INTO product_reviews VALUES (
2,
'The screen cracks easily, which is a drawback.',
'[0.9, 0.1, 0.1]'
);
COMMIT;
-- 3. Execute Vector Search (Similarity Search)
-- Find vectors close to the search query "Issues with durability" ([0.85, 0.15, 0.1])
-- This is a new syntax that does not exist in 19c
SELECT review_id, review_text
FROM product_reviews
ORDER BY vector_distance(embedding, '[0.85, 0.15, 0.1]', COSINE)
FETCH FIRST 1 ROWS ONLY;
② Developers Rejoice! Simplification of SQL (Better Developer Experience)
The “conventions” that were familiar up to 19c are no longer required.
-- [26ai] FROM DUAL becomes unnecessary
SELECT sysdate;
SELECT 1 + 1;
-- [26ai] BOOLEAN type can be used in table column definitions
CREATE TABLE user_settings (
user_id NUMBER,
is_active BOOLEAN DEFAULT TRUE
);
-- Insertion is also intuitive
INSERT INTO user_settings VALUES (101, FALSE);
INSERT INTO user_settings VALUES (102, TRUE);
-- Can be used directly in WHERE clauses (1=1 etc. is unnecessary)
SELECT * FROM user_settings WHERE is_active;
4. Migration and Operation Precautions (Troubleshooting)
We have listed points where you might get stuck when introducing 26ai.
- Enforcement of CDB/PDB Configuration
- Phenomenon: If you were using the “Non-CDB” configuration in 19c, you cannot upgrade as is.
- Solution: Using the AutoUpgrade tool automates the upgrade, including the conversion from Non-CDB to PDB. If performing this manually, you must migrate using Data Pump export/import (
expdp/impdp) withfull=yand set the import destination to a PDB.
- Client Compatibility
- Caution: Connections from very old Oracle Clients (such as 11gR2 or earlier) may not be supported.
- Recommendation: We strongly recommend updating the Oracle Client on the application server side to 19c or a 23ai/26ai client as well.
- Deprecated Features
- Verification: Old parameters such as
DB_UPGRADE_CACHE_SIZEand some old auditing features may be deleted or deprecated. Be sure to check the “Behavior Changes” section of the “Oracle Database 26ai Upgrade Guide” before upgrading.
- Verification: Old parameters such as
5. Frequently Asked Questions (FAQ)
Q1. Are 23ai and 26ai different things?
A1. Basically, they are the same code line. “Oracle AI Database 26ai” is the rebranded version that includes the features of Oracle Database 23ai plus the latest enhancements and a commitment to Long Term Support (LTS). Effort for application recertification will be minimal.
Q2. Can I upgrade directly from 19c?
A2. Yes, it is possible. Direct upgrades from Oracle 19c (19.x) to 26ai are supported. The use of the Oracle AutoUpgrade tool is recommended.
Q3. Can I use AI features in Standard Edition 2 (SE2)?
A3. Yes, many developer features, including AI Vector Search, are available in SE2. However, some options for parallel processing or performance enhancements (such as Exadata-specific features) may be limited to EE or Exadata environments.
6. Summary
Oracle AI Database 26ai is not just a version upgrade, but a platform that realizes “AI for Data.”
- Release: On-premises version planned for January 2026.
- AI Features: AI Vector Search completes RAG and similarity searches using only SQL.
- Verification: You can verify features for free immediately by using the Free Version.
- Migration: Direct migration from 19c is possible. However, moving to CDB/PDB is mandatory.
DBAs should start by launching a Free version container to check the new SQL syntax and the behavior of AutoUpgrade to prepare for the upcoming release.
Note
This article explains based on Oracle Database 19c and pre-release information for Oracle AI Database 26ai as of December 2025. Screens, default values, and detailed dates may change at the time of the actual product release. Please be sure to check the official documentation.
コメント