Database
  • The performance of re-scored binary quantization (AKA High Fidelity Quantized vector search) and scalar quantization is significantly improved. A re-scored binary reduces latency by approximately 5x and increases query throughput by 5x. Indexes built before v2026.08 that aren't regularly updated, must be re-created in order to get the full benefit. See Vector indexes - Cypher Manual.
🚀 New Cypher 25 features
  • Cypher now supports native full-text search through the new
    SEARCH
    clause for simple node and relationship patterns. See the Cypher Manual -> Query full-text indexes.
  • The functions
    toString()
    ,
    toStringList()
    and
    toStringOrNull()
    have been updated to accept
    LIST
    ,
    MAP
    ,
    NODE
    ,
    RELATIONSHIP
    and
    PATH
    types. See the Cypher Manual -> String functions.
  • Introduced map comprehension, which allows the creation of a
    MAP
    value by iterating over an existing
    MAP
    or
    LIST
    e.g
    {k: v IN map | keyExpression: valueExpression}
    . See the Cypher Manual -> Map comprehension.
  • Add
    AS COMMANDS
    for
    SHOW USERS
    and
    SHOW ROLES
    , to get a create role/user or grant role to user/auth rule command string that can be executed to recreate the current setup of roles, users and their assignments. See Operations Manual -> Manage users and Manage roles.
  • Add new
    AUTH RULES
    authRule
    filter to
    SHOW PRIVILEGES [AS COMMANDS]
    to filter on privileges assigned to roles granted to the given auth rules, similar to the existing
    USERS
    or
    ROLES
    filters. Also add some useful columns to the
    SHOW PRIVILEGES AS COMMANDS
    to be able to filter the commands on. See Operations Manual -> Role-based access control.
🛠️ Bug fixes
  • The page cache now reserves its full configured memory at startup and is guaranteed never to exceed
    server.memory.pagecache.size
  • $any
    dynamic type evaluation on a list containing invalid elements in the list tail will now correctly throw instead of short-circuit accept on the list head, and the error message now no longer erroneously suggests that
    LIST<LIST<STRING NOT NULL>>
    is an accepted outer list type.
  • A division by zero is now raising a clean division-by-zero arithmetic error instead of a previous overflow/out-of-range
For full details of all updates and fixes in this release, as well as links to relevant documentation please visit: Release Notes: Neo4j Aura Database – September 2026.