Short answer: AWS cut Glue prices 30% with version 6.0 — but the cut applies to the new version, not your current one. Every Glue job still pinned to an older version is now paying 30% more than it needs to for identical work. Upgrading requires no API changes, which makes this the rarest thing in cloud economics: a discount you claim by changing a version number.
Source: AWS Glue 6.0 now available with 30% lower price and full Apache Iceberg v3 support — AWS News Blog, August 21, 2026.
What Glue 6.0 includes
Glue 6.0 is generally available in all regions where Glue operates, on a modernized runtime: Apache Spark 4.1, Python 3.13, Scala 2.13. The headline items:
| Feature | What it changes |
|---|---|
| 30% lower price vs. previous Glue versions | Same work, cheaper hourly rate, still billed per second |
| Full Apache Iceberg v3 support (on Iceberg 1.11.0) | VARIANT with shredding, geo types, ns-timestamps, unknown-type resilience |
| Spark Declarative Pipelines | Declare transformations; the engine handles ordering and optimization |
| Arrow-native Python UDFs/UDTFs | Removes Python–JVM serialization overhead |
| Real-time streaming mode | Single-digit millisecond latency for stateless streaming |
The Iceberg v3 work is the most complete implementation on any fully serverless managed Spark service, per AWS. Three pieces stand out for practitioners: the VARIANT type with shredding speeds up queries over semi-structured data (query JSON and logs without flattening schemas first); geometry and geography types enable native spatial processing; and unknown-type handling lets pipelines survive upstream schema changes that would previously have failed the job.
The Data Catalog pricing is unchanged in structure: the first million stored objects and first million accesses per month are free.
Why the version pin is costing you
Glue jobs specify a version (--glue-version), and jobs don't upgrade
themselves. That's normally fine — but a 30% version-gated discount turns
every stale job into a standing overpayment. A data platform running
$3,000/month of Glue on an older version pays $900/month more than the
same jobs on 6.0, for nothing in return.
The upgrade friction AWS advertises is low: no API changes, select the
new version via the existing parameter in create-job/update-job APIs, or
%glue_version in notebooks, via Glue Studio, SDKs, CLI, or SageMaker
Unified Studio. For code compatibility, AWS ships a Spark upgrade agent
and an auto-upgrade feature for existing jobs.
What actually needs care is validation, not mechanics: jobs pinned to old versions usually stayed there because someone feared a Spark major-version jump (4.x here, from 3.x). The 30% saving funds a proper test cycle — run your production jobs against 6.0 on a subset, compare row counts and durations, then flip.
A quick prioritization order
- Dev and on-demand jobs first — cheapest to roll back, fastest to verify.
- Scheduled batch jobs with data-quality checks — the checks are your validation.
- Long-running streaming jobs last — the single-digit-millisecond streaming mode is new; test it separately.
Also verify your crawlers and interactive sessions while you're in there — they bill per second too, and they're the Glue spend that accumulates quietly outside the well-known ETL jobs.
The bigger pattern
Version-gated pricing is becoming a habit: infrastructure providers reserve their best rates for current-generation runtimes because it concentrates support load. The financial consequence is that doing nothing now has a price. A cost baseline that inventories which services run which versions — and what each version costs — turns "we should upgrade sometime" into a line item with a number on it.
This is a good example of the class of savings that only surfaces when someone watches continuously — price cuts land on services nobody remembers they run. Our cloud waste audit checklist covers the one-time sweep, and continuous cloud cost monitoring is how the next 30% off gets noticed the week it's announced rather than the quarter after.
FAQ
Does the 30% discount apply automatically to existing jobs?
No. It applies to jobs running Glue version 6.0. Jobs on older versions
keep their old pricing until you change --glue-version and re-verify.
Is upgrading risky? The runtime jumps from Spark 3.x to 4.1, so jobs with custom code need a validation pass. AWS provides a Spark upgrade agent and auto-upgrade tooling to smooth the code changes; the API surface itself is unchanged.
What is Iceberg v3's VARIANT type, in one sentence? A native type for semi-structured data that stores JSON-like values in a shredded, query-accelerated form — so you can query logs and documents without flattening them into columns first.