/

Data Integration

Best Snowflake ETL Tools in 2026

Intergalactic Data Labs

—

—

14 min read

Table of contents

Summarize

Who operates the pipeline and what it meters decide the best Snowflake ETL tool in 2026, since load paths have converged on staged COPY INTO or Snowpipe Streaming.

Method

Best for

Main limit

Engine you run

Typed tables on your own compute with no row meter

You own scheduling, upgrades, and alerts

Managed ELT service

Hundreds of sources and nothing to operate

Billed by rows, credits, events, or gigabytes

Snowflake native

Ingestion that never leaves Snowflake’s bill

Few SaaS sources, credit rates in a consumption table

Change data capture relay through Kafka

Teams that already run Kafka Connect

Raw event envelopes to flatten and a cluster to run

As of November 2026 twelve tools can run a production load into Snowflake, and this list ranks them on five facts from their own docs. Those are the pricing unit, log-based capture from Postgres, MySQL, and MariaDB, the load path, authentication now that Snowflake is retiring service passwords, and the LICENSE file. Load speed is left out, because no public benchmark covers most of these tools on a Snowflake route.

The 12 best Snowflake ETL tools at a glance

Tool

Pricing unit

Load path

Log-based capture

License

Documented limit

Filament sink in preview

None, self-hosted

PUT to internal stage, COPY INTO, MERGE

Postgres, MySQL

Apache 2.0

Unencrypted RSA key only, add-only schema changes

Fivetran

Monthly active rows

Internal stage, SQL merge

Postgres, MySQL, MariaDB

Proprietary

Hybrid deployment needs Enterprise

Airbyte

Credits on Cloud

Internal stage, typed final tables

Postgres, MySQL

ELv2

No MariaDB source, 5 minute floor on Cloud

Estuary Flow

GB plus per connector

Table stage, or Snowpipe Streaming for deltas

Postgres, MySQL, MariaDB

BSL 1.1

Delta tables are not fully reduced

Openflow

Snowflake compute credits

Snowpipe Streaming, MERGE from journal

Postgres, MySQL, MariaDB

Proprietary

Source auth by password only, not in trials

Snowpipe Streaming plus Kafka connector

Credits per uncompressed GB

Direct row writes, no stage

Needs a producer

Apache 2.0 connector

No source connectors of its own

Hevo

Events

Hevo’s S3 bucket, then load

Postgres, MySQL, MariaDB

Proprietary

30 minute minimum ingestion, no ARRAY columns

Matillion, now Maia

Credits by task hour

Avro to internal stage, merge

Postgres, MySQL

Proprietary

Capture only on the Scale edition, no price published

Stitch

Rows per month

Upsert loading

Postgres, MySQL, MariaDB

Proprietary

30 minute floor on every plan

dlt

None, library

PUT to per-table internal stage

Postgres

Apache 2.0

No scheduler, no MySQL log capture

Sling

Flat monthly token

Internal stage, native MERGE

Postgres, MySQL, MariaDB

GPL-3.0

Capture needs the 149 dollar Pro Max token

Debezium

None

Kafka, then the Snowflake connector

Postgres, MySQL, MariaDB

Apache 2.0

No Snowflake sink of its own

Galaxy builds Filament. Every other cell was read from the vendor’s own documentation, pricing page, or LICENSE file on November 19, 2026, and the order follows the five criteria above. Informatica, Talend, and Improvado appear in competing lists but were not verified for this one, so they are left out.

Fivetran, the managed catalog billed by active rows

Fivetran is the pick for the widest managed catalog if you accept a meter on rows. Its pricing page counts “the number of monthly active rows (MAR) processed by Fivetran in a calendar month”, with 500,000 MAR free and a 5 dollar base charge per connection below 1 million. The Postgres connector reads pgoutput, the MariaDB connector reads a row-format binary log, and Enterprise plans sync every minute.

The limits sit on the destination. The setup guide says you “must configure key-pair authentication for SERVICE users”, VARIANT values truncate at 128 MB, and Snowpipe Streaming is absent from the docs. Hybrid Deployment needs Enterprise or Business Critical, and the Postgres docs warn that the Query-Based fallback “may scan full tables to detect changes”.

Airbyte, the largest open catalog with a free Core

Airbyte wins on breadth you can self-host for nothing. Its pricing page says “Airbyte Core is free to self-host” and Cloud starts at 20 dollars a month for 5 credits. The Snowflake destination loads through the internal stage and has written typed final tables since version 4.0.0, with hard or soft deletes for captured changes.

The license and the floor are the limits. The LICENSE file is the Elastic License 2.0, free to use and modify but not open source. Cloud syncs no faster than hourly on Standard, 15 minutes on Plus, and 5 minutes on Pro, the registry holds no MariaDB source, and version 5.0.0 deprecated passwords on September 2, 2026.

Estuary Flow, continuous capture with Snowpipe Streaming

Estuary is the pick for seconds-fresh deltas from a managed service. Its Snowflake materialization states that “Snowpipe Streaming is additionally available for delta update bindings”, standard bindings merge from a table stage, and the schedule accepts a base interval of zero. Pricing is 0.50 dollars per gigabyte plus 100 dollars a month per connector, with 10 GB a month free.

The license is the limit. LICENSE-BSL is Business Source License 1.1 until November 20, 2029, and the docs warn that delta update tables “won’t be fully reduced”, so the fastest path trades a clean table for latency.

Snowflake Openflow, native ingestion built on Apache NiFi

Openflow is the pick when ingestion must stay inside the Snowflake contract. Its about page says it is “Built on Apache NiFi” and generally available in your cloud or inside Snowflake. The Postgres connector “loads data with Snowpipe Streaming” before a MERGE applies each change, and the MySQL connector covers MySQL 8 and MariaDB 11.4 or later.

The limits are operational. Source auth is username and password only, trial accounts need the account team to enable it, and the cost page says the management pool “continues to run and incurs costs, even if there are no runtimes running”. There is no Stripe connector and HubSpot is in preview.

Snowpipe Streaming and the Snowflake Connector for Kafka

This is the fastest load path Snowflake sells, and it is a load API rather than an ETL tool, so you bring the producer. The overview claims “As low as 5 seconds ingest-to-queryable latency”, and Kafka connector version 4 is generally available under Apache 2.0 with exactly-once ordered delivery.

The limit is everything upstream. The cost page bills “the volume of uncompressed data ingested” rather than warehouse time, but there are no source connectors, the classic connector lands records as two VARIANT columns, and you run Kafka Connect yourself.

Hevo, events pricing with a fixed-price enterprise tier

Hevo is the pick for a fixed bill at high volume. Its pricing page counts each changed record as one event, starts at 299 dollars a month for 5 million events, and sells Enterprise as “One fixed price for unlimited data volume per pipeline”. The Postgres source defaults to logical replication.

Three limits are documented. The frequency docs set a 30 minute minimum ingestion interval, data stages in Hevo’s own S3 bucket, and the Snowflake destination does not support ARRAY columns and warns that password logins “may stop connecting at any time”.

Matillion, now Maia, credits for pushdown ELT

Matillion, which became Maia and moved its docs to docs.maia.ai in the week of November 16, 2026, is the pick for SQL pushdown transformation with a capture add-on. Pricing is credits consumed by task hours with no public rate. The Snowflake streaming destination writes Avro to an internal stage and merges from there.

The limits are the gate and the topology. The streaming pipelines page states that “Streaming pipelines use Change Data Capture (CDC) and only currently support relational databases as sources”. Capture needs the Scale edition, each pipeline needs its own runner, and MariaDB is not a streaming source.

Stitch, row bands and a 30 minute floor

Stitch is the pick for the simplest meter. Its pricing page sells Standard from 100 dollars a month for 5 million rows and Advanced at 1,500 dollars for 100 million. The Postgres, MySQL, and MariaDB integrations all list log-based incremental replication, and the Snowflake destination upserts by default.

The floor is the limit. The scheduling docs offer intervals from 30 minutes to 24 hours, cron only on Advanced and Premium, there is no free tier, and key-pair auth appears only as a note in the setup form.

dlt, the Apache 2.0 Python library

dlt is the pick for a team that would rather own a Python repo than a vendor contract. The LICENSE file is Apache 2.0, and the pricing page promises “The dlt library stays Apache 2.0, always”. The Snowflake destination uses PUT and per-table internal stages, supports every write disposition, and takes password, key pair, OAuth, and the Snowpark Container Services token.

The limits are what a library leaves out. There is no scheduler or UI, the managed dltHub runtime costs 12,000 dollars a month, and only the pg_replication source captures from a log, so MySQL and MariaDB are batch sources.

Sling, a single binary with flat pricing

Sling is the pick for flat-priced capture from one binary. The CLI Pro page lists Pro at 79 dollars a month and Pro Max at 149 dollars with unlimited volume. The change capture docs list the Postgres write-ahead log and the MySQL and MariaDB binary logs, with changes “staged and applied with a native MERGE” into Snowflake.

The limits are the paywall and the license. Log capture requires Pro Max, the LICENSE file is GPL version 3, and the connection docs state that “Loading through an external stage fails, because Snowflake does not allow PUT on an external stage.”

Debezium plus the Snowflake Connector for Kafka

Debezium is the reference open source capture engine and the pick when Kafka already runs, because it has no Snowflake sink. The Debezium Server docs list Kafka, Kinesis, Pub/Sub, Pulsar, Redis, NATS, and Iceberg sinks, so the documented route is Kafka and then the connector above. The LICENSE is Apache 2.0, and 3.7.1.Final shipped on November 6, 2026 per the release notes.

The limit is the operating burden. You run Kafka Connect, the change envelope lands as VARIANT until you flatten it, and the Postgres and MariaDB connectors have no SaaS siblings for Stripe or HubSpot.

What each paid option bills

Ten of the twelve tools carry a price, and on top of each one Snowflake’s compute docs bill every warehouse resume for at least one minute.

Tool

Unit

Entry price

Free tier

Fastest sync and plan

Fivetran

Monthly active rows

5 dollar base per connection

500,000 MAR a month

15 min Standard, 1 min Enterprise

Airbyte Cloud

Credits

20 dollars a month, 5 credits

Core self-hosted

1 hour Standard, 15 min Plus, 5 min Pro

Estuary

GB plus connector

0.50 dollars per GB plus 100 dollars per connector

10 GB and 2 connectors

Base interval of zero

Hevo

Events

299 dollars a month, 5M events

14 day trial

30 min minimum ingestion

Matillion, now Maia

Credits by task hour

Not published

Trial

Continuous, Scale edition

Stitch

Rows per month

100 dollars a month, 5M rows

14 day trial

30 min on every plan

Openflow

vCPU or compute pool credits

Consumption table rate

None, not in trials

Continuous

Snowpipe Streaming

Credits per uncompressed GB

Consumption table rate

None

Continuous

Sling Pro

Flat monthly token

79 dollars, 149 for capture

GPL CLI without Pro

Continuous mode

dltHub

Credits

12,000 dollars a month, 5,000 credits

Library, 14 day trial

User scheduled

Which ETL tools run Postgres, MySQL, and Stripe to Snowflake?

The best ETL tools for PostgreSQL to Snowflake are the ones that read the write-ahead log, and every tool below does, while MariaDB and Stripe thin the field. Each column has a route guide, for Postgres, MySQL, MariaDB, Stripe, and HubSpot.

Tool

Postgres

MySQL

MariaDB

Stripe

HubSpot

Fivetran

Log

Log

Log

Events

Incremental

Airbyte

Log

Log

No source

Incremental

Incremental

Estuary Flow

Log

Log

Log

Real time

Real time

Openflow

Log

Log

Log

No connector

Preview

Hevo

Log

Log

Log

Incremental

Incremental

Matillion, now Maia

Log

Log

No

Batch

Batch

Stitch

Log

Log

Log

Incremental

Incremental

dlt

Log

Batch

Batch

Incremental

Incremental

Sling

Log, Pro Max

Log, Pro Max

Log, Pro Max

Batch

Batch

Debezium

Log

Log

Log

No

No

The best ETL tools for MySQL to Snowflake are the nine with a binary log reader. The best ETL tools for Stripe to Snowflake split between Fivetran and Estuary, which follow the Stripe events feed, and the batch readers that poll objects.

Which Snowflake ETL tool fits which workload?

A managed service earns its meter when nobody is free to run a pipeline, and a free engine earns its upkeep when the row count would make that meter the largest line on the bill.

Workload

Pick

Deciding fact

Hundreds of sources, nobody on call for pipelines

Fivetran

1 minute syncs on Enterprise, pgoutput and MariaDB binlog sources

Nothing outside the Snowflake contract

Openflow

Native, generally available, Snowpipe Streaming load, Postgres 11 and MariaDB 11.4

Seconds-fresh deltas from a managed service

Estuary Flow

Snowpipe Streaming on delta bindings, base interval of zero

Kafka already in production

Debezium plus the Kafka connector

Apache 2.0 end to end, 5 to 10 second latency per Snowflake

One Python repo and no vendor

dlt

Apache 2.0 library, the widest Snowflake auth list, pg_replication

Flat price with MariaDB capture

Sling Pro Max

149 dollars a month, native MERGE into Snowflake

Fixed bill at high volume

Hevo Enterprise

One price per pipeline, unlimited events

Self-hosted UI with the widest catalog

Airbyte Core

Free under ELv2, typed tables since 4.0.0

The Best CDC Tools guide ranks the capture engines on their own, the LICENSE audit checks the open source rows, and data replication vs ETL explains why most of these tools are loaders.

Frequently asked questions

What is the best ETL tool for Snowflake?

There is no single best tool, because the right one depends on who operates the pipeline and what it meters. Fivetran suits teams that want managed connectors billed by active rows, Openflow suits teams that want nothing outside Snowflake, and dlt, Sling, or Debezium suit teams running their own engine.

Does Snowflake have its own ETL tool?

Yes. Openflow is a generally available ingestion service built on Apache NiFi that runs in your cloud or inside Snowflake and loads with Snowpipe Streaming. Snowpipe and Snowpipe Streaming are the load paths underneath it, and COPY INTO is the bulk command every third-party tool calls through a stage.

What is the difference between ETL and ELT in Snowflake?

ETL transforms rows before they land, while ELT loads raw rows first and transforms them with SQL inside the warehouse. Snowflake bills warehouses per second with a one minute minimum, so ELT pays for transformation compute in Snowflake. Most tools on this list are loaders that leave transformation to SQL.

What are the Fivetran alternatives for Postgres to Snowflake change data capture?

Airbyte, Estuary Flow, Openflow, Hevo, Matillion, Stitch, Sling, dlt, and Debezium all document log-based capture from Postgres into Snowflake. Airbyte and Hevo bill by credits or events, Estuary by gigabyte, Openflow by Snowflake compute, Sling needs a Pro Max token, and dlt and Debezium are free engines you operate.

Which tools stream Postgres or MySQL changes to Snowflake?

Openflow reads the Postgres write-ahead log or the MySQL binary log and loads through Snowpipe Streaming. Estuary Flow captures continuously and can push deltas through Snowpipe Streaming too. Debezium streams both databases into Kafka, where the Snowflake Connector for Kafka lands rows in 5 to 10 seconds, and Sling Pro Max runs both logs continuously.

How do I automatically sync MySQL to Snowflake?

Turn on row-format binary logging, create a Snowflake service user with a key pair, and point a tool with a MySQL binlog source at both ends. Fivetran, Airbyte, Estuary, Openflow, Hevo, Stitch, and Sling all document that path and schedule the sync once the connection validates.

Can open source tools handle Snowflake ETL?

Yes, with different licenses. dlt and Debezium are Apache 2.0, Sling is GPL-3.0, and Airbyte Core is free to self-host under the Elastic License 2.0, which the Open Source Initiative does not count as open source. All four load through an internal stage or Kafka and leave scheduling and monitoring to you.

Why is my Snowflake ETL tool failing to log in with a password?

Because Snowflake is disallowing passwords for service users on a rolling per-account basis between August and October 2026. Once your account is enforced, any destination configured with a username and password stops connecting. Fivetran, Airbyte, Estuary, and Hevo all tell customers to switch to key-pair authentication or a programmatic access token.

What is the most cost-predictable Snowflake ETL pricing model?

A flat fee. Sling Pro Max is 149 dollars a month with unlimited volume, Hevo’s Enterprise tier is one fixed price per pipeline, and Stitch sells row bands from 100 dollars. Active rows, credits, and gigabytes all move with your data, and Snowflake warehouse credits move with how often a tool wakes the warehouse.

Do I need Snowpipe if I use an ETL tool?

No. Snowpipe and Snowpipe Streaming are load paths, not pipelines. Every third-party tool already lands data through a stage and COPY INTO or through Snowpipe Streaming. You touch Snowpipe directly only when you write your own files to a stage or your own producer to the streaming API, which Openflow and the Kafka connector do for you.

More articles

Stay up to date with what we’re building

Stay up to date with what we’re building

Stay up to date with what we’re building

Questions

Answered

FAQ

What does Galaxy do?

What is Filament?

What is enterprise context management?

What does working with Galaxy look like?

How do you handle security and compliance?

Why does Galaxy build in the open?

Own your knowledge stack

Own your knowledge stack

Copyright © 2026 Galaxy. All rights reserved.