2026 is a year for some major changes in the MonetDB database engine, which the core team has been working on for many months. Due to the significance of these changes, only one feature release has been planned. In addition, the MonetDB team will continue working on improving the robustness of the most recent major release Dec2025.
A major release is planned in September 2026, which will contain the following main new features:
MonetDB’s column-at-a-time query execution engine will be extended with a pipelined engine that (logically) divides the data of a column into multiple partitions and processes the data blocks in parallel through pipelines formed by the relational operators of an SQL query.
This parallel pipeline engine can significantly boost the performance of some relational operators, such as GROUP BY, LIMIT and joins. Moreover, this new engine enables fine-grained control of hardware resource consumption (e.g. CPU and memory) per user, query, etc.
The new block-at-a-time data processing scheme will cooperate with MonetDB’s existing column-at-a-time processing scheme. New query optimisers will be added to determine the best processing scheme of each relational operator based on the properties of the data the operator operates on.
Not only more general purpose statistics, but also more advanced statistics, such as multi-column group counts, will be added to MonetDB. The later is crucial for the pipeline engine to make decisions earlier in the query plan generation about, for instance, what algorithm to use, if and how to partition the data and create hash table, and if a perfect hash is needed.
Many data processing systems (including the current MonetDB versions) simply store nested data (e.g. JSON data) as a big string. Although this strategy is relatively easy to implement, queries on those data are often inefficient and impossible to optimise.
In this feature release, support for nested data as a native data type will be added. The input nested data is disassembled into different types and stored in corresponding columns. For example, a JSON document containing the name, date of birth, age and salary information of persons is stored as a table of four columns of the type VARCHAR, DATE, INT and DOUBLE, respectively.
With the native support, queries on native data automatically benefit from any existing and future optimisation techniques in MonetDB, which can result in orders of magnitude of performance improvement.
We expect to produce 2 or 3 bugfix releases for the current MonetDB version Dec2025 before the next major version is released, although there are no fixed dates yet for these minor releases.
Further, after the above-mentioned major release, we’ll probably make improvements upon that version with one bugfix release within this year.
A parser for vector data has been added. In 2026, we’ll continue to work on an efficient storage scheme for AI workloads, i.e. it should be both space efficient and supportive for the kNN algorithms. The basic principle is similar to that described in PDX, combined with our idea of dynamic numerical columns, we’re working on supporting sub-byte data types, e.g. 2-bits FLOATs. This storage optimisation allows us to minimise memory usage while grouping correlated dimensions together to speed up kNN search queries.
We expect to finish much of the above work later in 2026, then, we’ll be able to make release plan for it.
The items on our wish list are largely unchanged. For more information about them, please see the earlier roadmaps. In 2026, we’ll focus on the release of the big new features. After that is done, probably in 2027, we shall reconsider the wish list for the next steps.
Dynamic numerical columns
Encrypted column storage
Time series support
Graph support