Quantrail Data

ClickHouse® 25.12, released on December 18, 2025, is a major stability and performance-focused release that strengthens ClickHouse’s position as a production-ready analytics platform.

Rather than introducing flashy surface-level features, 25.12 delivers deep engine improvements: faster JOINs, smarter query planning, safer schema changes, better vector handling, and meaningful operational upgrades for cloud, lakehouse, and high-ingest workloads.

This article breaks down the most important official changes – and explains why they matter in real systems.

Key Highlights in ClickHouse® 25.12

1. JOINs & Query Planning: Faster, Smarter, More Predictable

ClickHouse® 25.12 continues heavy investment in JOIN performance and correctness.

What’s new

• JOIN runtime filters now support ANTI JOINs, reducing unnecessary data scans.
• JOIN reordering improvements using a new experimental DPsize algorithm, helping the planner choose more efficient join orders for complex INNER JOIN queries.
• Improved read-in-order behavior across JOINs, allowing downstream stages to reuse ordering and reduce sorting costs.
• Lower memory usage for JOINs with additional filters, fixing regressions seen in previous versions.

Why this matters

JOIN-heavy analytics is often where ClickHouse® clusters hit:

• memory pressure
• latency spikes
• unpredictable query plans

25.12 makes JOIN execution more stable and easier to reason about at scale.

2. Performance Improvements Across the Engine

This release includes dozens of low-level performance optimizations, many of which stack together in real workloads.

Notable improvements

• ORDER BY … LIMIT N optimizations using skip indexes and dynamic thresholds.
• Skip indexes now work with mixed AND/OR conditions in WHERE clauses (enabled by default).
• Faster index analysis when large minmax indexes are present.
• Improved lazy column materialization, reducing CPU and memory overhead.
• Better SIMD usage for logical functions, sorting, and boolean conversions.
• Improved Decimal comparison performance.
• Optimized MergeTree reader filtering and merge behavior.

Why this matters

These changes don’t require query rewrites – existing queries simply get faster, especially on large datasets.

3. Vector & QBit Improvements: Safer and More Robust

Vector workloads continue to mature in ClickHouse® 25.12.

Key updates

• Implicit type casting from Array → QBit is now supported, simplifying inserts.
• Multiple crash and correctness fixes for L2DistanceTransposed, including:
  • non-constant reference vectors
  • mismatched element types
  • invalid precision values
• Vector index cache entries are cleaned up immediately when table parts are dropped.
• Additional safety checks prevent malformed QBit data from crashing the server.

Why this matters

ClickHouse® is becoming increasingly viable for:

• similarity search
• embedding analytics
• ML-assisted querying

25.12 focuses on correctness, safety, and operational reliability for vector workloads.

4. SQL & Language Enhancements

ClickHouse® SQL continues to become more expressive and intuitive.

New and improved SQL features

• Non-constant arguments are now allowed in IN, including tuple-based inputs.
• Negative tuple indexing is supported (e.g. tuple.-1).
• INSERT into ALIAS columns is now allowed.
• New cryptographic HMAC() function.
• Geometry functions for area and perimeter calculations.
• Improved windowFunnel behavior with allow_reentry, enabling more accurate funnel analytics.

Why this matters

Cleaner SQL means:

• fewer workarounds
• simpler analytics logic
• more maintainable pipelines

5. Storage, Lakehouse & Cloud Integrations

ClickHouse® 25.12 strengthens integration with modern data platforms.

Highlights

• Iceberg support improvements, including ORDER BY support and safeguards against large partition counts.
• Delta Lake Change Data Feed (CDF) support via table functions.
• S3 / Azure Queue enhancements, including file tagging and move actions after processing.
• Improved S3 performance by pushing down path filters and avoiding unnecessary listings.
• Pre-signed S3 URLs are now fully supported.

Why this matters

These changes reduce operational friction when ClickHouse® is used as part of:

• lakehouse architectures
• object-storage–based analytics
• cloud-native pipelines

6. Backward-Incompatible & Behavioural Changes (Important!)

Some changes in 25.12 require attention before upgrading.

Key breaking changes

• ALTER MODIFY COLUMN now requires an explicit DEFAULT when converting Nullable → non-nullable.
• Non-comparable types are no longer allowed in ORDER BY or comparison functions.
• Special MergeTree engines can no longer be created with empty ORDER BY keys.
• Projection positional arguments are now disabled by default (opt-in available).
• Ngram tokenizer behavior changed – empty tokens now return no rows.

Upgrade tip

Run upgrades in staging first and review schema mutations and text search behavior carefully.

7. Observability, Operations & Safety

Operational visibility improves significantly in 25.12.

What’s new

• New system tables for background job introspection
• Better error logging, including last error metadata
• Improved ClickHouse client exit codes, making automation safer
• Improved Keeper and ZooKeeper compatibility
• Reduced memory usage during merges and background tasks

Why this matters

These changes make ClickHouse® easier to operate, debug, and automate in production.

Why ClickHouse® 25.12 Is a Meaningful Release

This is a “hardening” release.

ClickHouse® 25.12 focuses on:

• performance consistency
• correctness under edge cases
• safer upgrades
• real-world production behavior

If you’re running:

• JOIN-heavy analytics
• large MergeTree tables
• vector similarity workloads
• cloud or lakehouse pipelines

25.12 is a high-value upgrade.

Final Thoughts

ClickHouse® 25.12 doesn’t chase headlines – it improves the foundation.

With faster JOINs, smarter query planning, stronger vector reliability, and better operational tooling, this release makes ClickHouse® more dependable at scale.

For teams already using ClickHouse® in production, the gains are real and measurable.

Need Help Upgrading or Optimizing ClickHouse®?

At Quantrail Data, we help teams:

• Upgrade ClickHouse safely
• Optimize JOIN-heavy workloads
• Design analytics + vector pipelines
• Operate ClickHouse reliably at scale

If you’re planning a 25.12 upgrade or want expert guidance, we’re happy to help.

References

ClickHouse Release 25.12 – Official Changelog
ClickHouse Release 25.11 – blog