There is no single best ETL tool, because the category covers different jobs. To copy data from SaaS apps and databases into a warehouse without running anything yourself, look at Fivetran or Hevo Data. To self-host connectors, Airbyte (source-available) or the open source Meltano and dlt. For large organisations that need data quality, governance and on-premises reach in one contract, Informatica or Qlik Talend Cloud; for low-code pipelines on a cloud warehouse, Matillion. To schedule and monitor everything, use an orchestrator: Apache Airflow, Dagster or Prefect. If you are committed to one cloud, start with its own service (AWS Glue, Fabric Data Factory, Google Cloud Dataflow and Datastream). For SQL transformations inside the warehouse, dbt. Many teams combine one ingestion tool, dbt and an orchestrator.
"ETL tool" is used for at least five kinds of product, and comparing across kinds causes most bad buying decisions. Ingestion (EL) tools such as Fivetran and Airbyte extract data from sources and load it into a warehouse or lake, usually with little transformation. Transformation tools such as dbt run SQL inside the warehouse after loading, which is the ELT pattern. Integration suites such as Informatica and Qlik Talend Cloud bundle ingestion, transformation, data quality and governance. Orchestrators such as Airflow, Dagster and Prefect do not move data themselves; they schedule, retry and monitor the steps that do. Cloud-native services from AWS, Microsoft and Google cover several of these jobs inside one cloud.
If the difference between ETL and ELT is new to you, read ETL vs ELT and ETL vs data pipeline first. For individual decisions, see Airbyte vs Fivetran, Apache Airflow vs Airbyte, Apache Airflow vs Dagster and Talend vs Informatica. If you are moving away from Fivetran specifically, see the best Fivetran alternatives.
Quick picks
Managed connectors with SaaS or Hybrid Deployment, a Free plan of 500,000 MAR a month, and dbt now part of the same company.
Free self-managed Core edition (source-available, ELv2) plus a cloud service with credit or capacity pricing.
An Apache 2.0 Python library that infers schemas and loads incrementally, with no server to run.
Apache-licensed, with managed versions from AWS, Google and Astronomer, and Airflow jobs in Fabric.
Version-controlled, tested SQL models run inside your warehouse; the open source core is free.
Pipelines, copy jobs, mirroring and Dataflow Gen2 inside Fabric, with Azure Data Factory still sold pay as you go.
How we chose
We considered managed ELT services (Fivetran, Hevo Data, Stitch, Estuary), open source and source-available ingestion (Airbyte, Meltano, dlt), enterprise suites (Informatica, Qlik Talend Cloud, Matillion), orchestrators (Apache Airflow, Dagster, Prefect), cloud-native services (AWS Glue, Azure Data Factory and Fabric Data Factory, Google Cloud Dataflow and Datastream) and dbt for transformation. To be included, a product had to be sold or actively developed in October 2026, publish its pricing model or licence on an official page, and be a realistic choice for loading data into a SQL warehouse or database.
Stitch is not listed as an entry. Its site now says "Stitch is now part of Qlik", encourages new users to try Qlik Talend Cloud instead, and its trial opens Qlik Talend Cloud. Plans are still listed for existing customers and we found no end-of-life date, but we do not recommend it for new projects (see Fivetran vs Stitch). Talend Open Studio, the former free edition, was retired on 31 January 2024 according to Qlik. Estuary is covered in the Fivetran alternatives guide.
For each product we recorded its type, pricing model, licence, hosting options, free plan or trial, and main trade-off. Where a price is quoted it is one dated example from the vendor's own page; we do not estimate anyone's bill. The order is not a ranking: entries are grouped by type, in the order managed ELT, open source ingestion, enterprise suites, orchestration, cloud-native services and transformation.
At a glance
| Tool | Type | Pricing model | Open source | Hosting | Best for | Main trade-off |
|---|---|---|---|---|---|---|
| Fivetran | Managed ELT | Monthly Active Rows per connection; Free plan | No (proprietary) | SaaS or Hybrid Deployment | Hands-off ingestion from many sources | Usage costs are hard to forecast |
| Hevo Data | Managed ELT | Events per month; Free plan | No | SaaS | Smaller teams wanting a simple managed tool | No self-hosted option |
| Airbyte | Ingestion platform | Free self-managed; cloud credits or capacity | Source-available (ELv2) | Self-managed, cloud or in your own boundary | Self-hosting and custom connectors | You run and upgrade it yourself on Core |
| Meltano | Ingestion framework (CLI) | Free; paid hosting and support from Matatika | Yes (MIT) | Self-hosted or managed | Engineers who want pipelines as code | More engineering effort than a managed service |
| dlt | Python ingestion library | Free library; paid dltHub platform | Yes (Apache 2.0) | Runs wherever Python runs | Custom API and database loads in Python | Needs Python skills and somewhere to run it |
| Informatica Cloud Data Integration | Enterprise suite | Informatica Processing Units (IPUs) | No | SaaS on several clouds | Large enterprises needing governance and MDM | Enterprise pricing and complexity |
| Qlik Talend Cloud | Enterprise suite | Capacity by data volume, job runs and duration | No (open edition retired) | SaaS or client-managed | Integration plus data quality in one contract | No public prices |
| Matillion Maia Foundation | Low-code ELT | Credits consumed by task hours and extra users | No | SaaS; hybrid agents on Scale | Low-code pipelines on a cloud warehouse | No public starting price |
| Apache Airflow | Orchestrator | Free; managed services billed separately | Yes (Apache 2.0) | Self-hosted or managed (MWAA, Google, Astronomer) | Scheduling and monitoring many tools | Does not move data by itself |
| Dagster | Orchestrator | Free OSS; Dagster+ credits per materialisation | Yes (Apache 2.0) | Self-hosted, serverless or hybrid | Asset-centred pipelines with lineage | Different model to learn than task DAGs |
| Prefect | Orchestrator | Free OSS; Prefect Cloud plans | Yes (Apache 2.0) | Self-hosted or Prefect Cloud | Python teams wanting light orchestration | Smaller provider ecosystem than Airflow (editorial) |
| AWS Glue | Cloud-native ETL | DPU-hours billed per second | No | AWS only | Spark ETL and cataloguing on AWS | AWS only; Spark knowledge helps |
| Fabric Data Factory and Azure Data Factory | Cloud-native ETL and orchestration | Fabric capacity; ADF pay as you go | No | Microsoft cloud; on-premises via gateway or runtime | Microsoft and Power BI shops | Some ADF features not yet in Fabric |
| Google Cloud Dataflow and Datastream | Cloud-native streaming and CDC | Dataflow: vCPU, memory and engine usage | Beam is open source; services are not | Google Cloud | Streaming and CDC into BigQuery | Google Cloud centred |
| dbt | Transformation (in-warehouse SQL) | Free core; dbt platform per seat and models built | Yes for dbt Core (Apache 2.0) | Self-run or dbt platform | Tested, version-controlled SQL models | Transforms only; needs an ingestion tool |
Fivetran
Fivetran is a managed ELT service: you configure connections and it extracts and loads data into your warehouse or lake, handling schema changes and incremental syncs. It documents two deployment models, SaaS and Hybrid Deployment, where data is processed inside your own network while Fivetran's cloud orchestrates it, and a Connector SDK for sources it does not cover.
Billing is by Monthly Active Rows (MAR): distinct rows inserted or updated in a month, counted by primary key. Since 1 March 2025 Fivetran calculates its tiered rates per connection rather than across the account; its FAQ says users with many similar-sized connections will pay more and those with one dominant connection less. The Free plan includes 500,000 MAR and 5,000 model runs a month, and Fivetran's documentation lists a USD 5 base charge for each standard connection with between 1 and 1 million MAR on paid plans (October 2026). Paid plans are Standard, Enterprise and Business Critical.
Corporate status: Fivetran and dbt Labs completed an all-stock merger on 1 June 2026 and operate as "Fivetran + dbt Labs". Fivetran also announced an agreement to acquire Census (reverse ETL) in May 2025, and its pricing now includes Activations. See Airbyte vs Fivetran and Fivetran vs Matillion.
- Costs scale with changed rows per connection, so bills are hard to forecast for high-churn tables.
- Proprietary and not self-hostable; Hybrid Deployment still depends on Fivetran's cloud for control.
- Mainly loads data; heavier transformations need dbt or another tool.
Hevo Data
Hevo Data is a managed, no-code pipeline service. It bills by events, where each record inserted, updated or deleted in the destination counts as one event. The Free plan covers up to 1 million events a month with limited connectors and up to five users. As listed on Hevo's pricing page on 7 October 2026, the Starter plan begins at USD 265 a month billed annually (USD 299 monthly), with Professional and Business Critical tiers above it and a 14-day trial with no card.
Starter includes dbt integration, and Business Critical adds streaming pipelines, SSO and VPC peering. See Fivetran vs Hevo Data and Airbyte vs Hevo Data.
- Cloud only; Hevo does not list a self-hosted or open source edition.
- The Free plan has limited connectors and hourly scheduling.
- Features such as SSO and streaming pipelines are reserved for the custom-priced tier.
Airbyte
Airbyte is a data movement platform with a large connector catalogue and a connector builder. Airbyte Core is the free, self-managed edition. Airbyte's licence page states that its connectors and public repositories, except the Airbyte Protocol (MIT), are under the Elastic License 2.0 (ELv2). ELv2 is source-available rather than an OSI-approved open source licence: you may run and modify it, including inside commercial products, but you may not offer Airbyte to others as a managed service.
The paid plans are Standard and Plus, priced by credits based on data volume, and Pro and Enterprise Flex, priced by capacity in "Data Workers"; Enterprise Flex runs inside your own boundary. As listed on Airbyte's pricing page on 7 October 2026, Standard starts at USD 20 a month, and paid tiers have a 14-day trial. See Airbyte vs Fivetran and Airbyte vs Meltano.
- On Core you host, monitor, scale and upgrade the platform yourself.
- ELv2 is source-available, not OSI open source, and forbids reselling Airbyte as a service.
- Capacity pricing on Pro and Enterprise Flex is by quote.
Meltano
Meltano describes itself as a declarative, code-first data integration engine. It is MIT-licensed on GitHub and runs as a command-line tool, with pipelines defined in project files that live in Git. Its site describes it as "an open source project by Matatika", which also offers a managed option and paid support.
It suits teams that want ingestion versioned and reviewed like application code, and who are comfortable running it in their own scheduler or orchestrator. See Airbyte vs Meltano.
- Command-line and configuration based, so it needs engineering time rather than a point-and-click UI.
- Connector quality varies, so test each extractor you depend on (editorial).
- You are responsible for hosting, scheduling and monitoring unless you buy the managed option.
dlt
dlt (data load tool) is an Apache 2.0 Python library installed with pip. You write a Python function that yields data and dlt infers the schema, normalises nested data, manages incremental state and loads the result into a destination. There is no backend service to run, so a pipeline can run in a notebook, a cron job, a serverless function or an orchestrator task.
dltHub, the company behind it, sells a managed platform that adds a runtime, observability and data quality checks. dlt suits teams whose sources are internal APIs or awkward systems that packaged connectors do not cover.
- You write and maintain Python code for each source.
- Scheduling, retries and alerting come from whatever you run it in.
- dltHub platform pricing was not clearly published when checked.
Informatica Cloud Data Integration
Informatica's current platform is Intelligent Data Management Cloud (IDMC), which includes Cloud Data Integration, Application Integration, Data Quality and Observability, Data Catalog, governance and master data management. Usage is metered in Informatica Processing Units (IPUs). Informatica offers a 30-day cloud trial and a free Cloud Data Integration tier.
Salesforce completed its acquisition of Informatica on 18 November 2025, and Informatica's site now says it is part of Salesforce. Informatica states it remains multi-cloud. See Talend vs Informatica.
- Enterprise pricing and procurement; IPU consumption needs to be planned with the vendor.
- A broad platform with many products to learn, more than a small team usually needs.
- Ownership changed in November 2025, so confirm product and support commitments in your contract (editorial).
Qlik Talend Cloud
Qlik acquired Talend in 2023 and now sells Qlik Talend Cloud in Starter, Standard, Premium and Enterprise editions. Qlik says usage is measured by a combination of data volume moved, number of job executions and execution duration, and customers subscribe to a capacity level; prices are not published. Qlik also offers Qlik Talend as a client-managed solution, and the commercial Talend Studio is still sold.
The free Talend Open Studio was retired on 31 January 2024 and is no longer hosted or updated, according to Qlik. Stitch, the managed ELT service that came with Talend, now points new users to Qlik Talend Cloud. See Fivetran vs Talend, Matillion vs Talend and Airbyte vs Talend.
- No public pricing; every edition is sold through Qlik sales.
- There is no longer a free open source Talend edition.
- Customers of older Talend and Stitch products face migration work onto Qlik Talend Cloud.
Matillion Maia Foundation
Matillion's current product is Maia Foundation (formerly the Data Productivity Cloud), part of its Maia platform, which includes Maia, which Matillion calls its agentic data workforce. Pipelines are built on a low-code canvas with SQL and Python and pre-built connectors, with built-in Git. Plans are Developer (one developer user), Teams and Scale (five developer users each); credits are consumed by task hours and by developer users beyond those included.
Scale adds hybrid deployment with self-hosted agents, custom SSO, lineage and streaming change data capture. Matillion is sold through the AWS, Azure and Snowflake marketplaces. See Fivetran vs Matillion and Airbyte vs Matillion.
- No starting price on the public pricing page.
- Hybrid deployment and streaming CDC are limited to the Scale plan.
- Proprietary pipelines are harder to move to another tool than plain SQL or Python (editorial).
Apache Airflow
Apache Airflow is an Apache Software Foundation orchestrator in which workflows are Python DAGs. Airflow 3.0 was released on 22 April 2025 with DAG versioning, scheduler-managed backfills, a Task Execution Interface that allows tasks to run in other environments, event-driven scheduling and a new React UI. The latest release when checked was 3.3.2 (17 September 2026). The Airflow project lists Airflow 2 as end of life since 22 April 2026, although managed services may still run 2.x builds for a period.
Managed options: Amazon MWAA supports Airflow up to 3.3.1; Google's Cloud Composer, now being renamed Managed Service for Apache Airflow, supports Airflow 3.3.1 in Gen 3; and Astronomer's Astro bills deployments and workers by usage. Fabric Data Factory can also run Airflow jobs. Airflow does not extract data by itself; it calls tools such as Airbyte, Fivetran or dbt. See Apache Airflow vs Airbyte, Apache Airflow vs Fivetran and Apache Airflow vs Prefect.
- Self-hosting means running a scheduler, metadata database, workers and upgrades.
- An orchestrator only; it needs operators or other tools to move and transform data.
- Airflow 2 is end of life, and moving to Airflow 3 needs code and configuration changes, so plan the upgrade.
Dagster
Dagster is an Apache 2.0 orchestrator that models pipelines as data assets declared in Python, with lineage and observability built in. Dagster+ is the managed service, in Solo, Starter and Pro plans, with serverless or hybrid deployment (ECS, Kubernetes, Docker and others). Usage is measured in credits, where each asset materialisation or op execution counts as one credit; as listed on Dagster's pricing page on 7 October 2026, Starter is USD 100 a month, with a 30-day trial.
See Apache Airflow vs Dagster and Prefect vs Dagster.
- The asset-based model takes time to learn for teams used to task DAGs.
- Credit-based billing grows with the number of materialisations.
- Fewer managed hosting choices than Airflow, which several clouds offer.
Prefect
Prefect is an Apache 2.0 Python orchestration framework, currently on major version 3, in which ordinary functions become flows and tasks through decorators, with retries, caching, scheduling and event triggers. You can self-host the server or use Prefect Cloud.
Prefect Cloud has a free Hobby plan (2 users, 5 deployments and 500 minutes of serverless compute a month, as listed on 7 October 2026), Starter, Team and custom-priced Enterprise plans. See Apache Airflow vs Prefect and Prefect vs Dagster.
- Requires Python 3.11 or later for current releases.
- SSO, RBAC and multiple workspaces are Enterprise features in Prefect Cloud.
- Fewer cloud-provider managed offerings than Airflow.
AWS Glue
AWS Glue is a serverless data integration service with ETL jobs, crawlers, interactive sessions and the Glue Data Catalog, which other AWS analytics services use. Jobs and crawlers are billed per second in Data Processing Unit hours; AWS listed USD 0.44 per DPU-hour on its pricing page on 7 October 2026, with rates varying by region. The first million Data Catalog objects stored and the first million requests are free.
AWS also offers zero-ETL integrations, for which it states there is no additional fee beyond the source and target resources used. Glue suits teams already standardised on AWS and on Amazon Redshift or S3-based lakes.
- AWS only, so it does not help with multi-cloud data movement.
- Custom jobs are Spark or Python code, which is more engineering than a managed connector.
- SaaS connector coverage is narrower than dedicated ELT services (editorial).
Fabric Data Factory and Azure Data Factory
Microsoft describes Data Factory in Microsoft Fabric as "the next generation of Azure Data Factory". It provides pipelines, copy jobs (bulk, incremental and change data capture copy), mirroring of operational databases into OneLake, Dataflow Gen2 for low-code transformation, dbt jobs and Apache Airflow jobs. It is billed through Fabric capacity (F SKUs), and Microsoft publishes an upgrade guide for Azure Data Factory and Synapse pipelines.
Azure Data Factory is still sold as a pay-as-you-go service, billed per activity run, per integration runtime hour for data movement and per vCore-hour for data flows, with a self-hosted integration runtime for on-premises sources. Microsoft's comparison lists Azure-SSIS integration runtimes and managed virtual networks with private endpoints as "to be determined" in Fabric, so ADF remains the option for those needs today.
- Some ADF capabilities, including SSIS package hosting, are not yet in Fabric.
- Fabric capacity is shared with other workloads, so heavy pipelines compete with reports and queries.
- Strongest within the Microsoft ecosystem; less natural for AWS or Google Cloud estates.
Google Cloud Dataflow and Datastream
Google Cloud splits this work across two services. Dataflow is a managed runner for Apache Beam pipelines, for batch and streaming, billed on worker vCPU and memory plus Shuffle or Streaming Engine usage. Datastream is a serverless change data capture and replication service that reads from MySQL, Oracle, PostgreSQL (including AlloyDB), SQL Server, MongoDB and Spanner, and from some SaaS applications such as Salesforce, and writes to BigQuery, Cloud Storage or Apache Iceberg tables.
Together they suit teams whose destination is BigQuery. For orchestration Google offers Managed Service for Apache Airflow (formerly Cloud Composer). See BigQuery alternatives if the destination itself is in question.
- Dataflow pipelines are Apache Beam code, which needs engineering skills.
- Datastream covers databases and a few SaaS sources, not a broad SaaS catalogue.
- Designed around Google Cloud destinations.
dbt
dbt is the "T" in ELT: you write SELECT statements as models, and dbt builds them as tables or views in your warehouse in dependency order, with tests and documentation. dbt Labs announced, on completing its merger with Fivetran on 1 June 2026, that dbt Core v2.0 open sources the dbt Fusion engine runtime under the Apache 2.0 licence.
The hosted dbt platform has a free Developer plan (one developer seat, 3,000 successful models built a month, one project), a Starter plan at USD 100 per user per month (as listed on 7 October 2026), and Enterprise plans. dbt does not extract or load data, so it is paired with an ingestion tool and often an orchestrator. Fabric also runs dbt jobs natively.
- Transformation only; it needs a separate ingestion tool.
- Platform pricing combines seats and models built, so cost grows with team size and model count.
- Now owned by the same company as Fivetran, which some buyers will weigh when choosing their ingestion tool (editorial).
How to choose
Start from the job, not the brand. If your problem is getting data out of SaaS apps and databases, choose an ingestion tool first: managed (Fivetran, Hevo Data) if you want no infrastructure and can accept usage-based bills, or self-hosted (Airbyte, Meltano, dlt) if you have engineers and want control over cost and data location. If your data already lands in the warehouse and the problem is messy SQL, add dbt. If you have several tools and scripts running on cron, add an orchestrator: Airflow for breadth and managed options, Dagster for an asset-first model, Prefect for light Python workflows.
Enterprise suites (Informatica, Qlik Talend Cloud) earn their cost when you also need data quality, governance, master data or on-premises reach under one contract. Cloud-native services (AWS Glue, Fabric Data Factory, Google Dataflow and Datastream) are often the cheapest route to try when everything lives in one cloud, because they are already in your bill and identity system. Whatever you pick, trial it with your own sources and data volumes, and use the vendor's calculator rather than list prices.
Frequently asked questions
What is the difference between ETL and ELT tools?
ETL tools transform data before loading it into the destination; ELT tools load raw data first and transform it inside the warehouse, usually with SQL. Most modern ingestion tools (Fivetran, Airbyte, Hevo Data) are ELT and pair with dbt. See ETL vs ELT.
Is there a free ETL tool?
Yes. Meltano (MIT), dlt (Apache 2.0), Apache Airflow, Dagster and Prefect (Apache 2.0) and dbt Core are free to self-host. Airbyte Core is free but source-available under ELv2. Fivetran, Hevo Data and the dbt platform have free plans with usage limits.
Is Airflow an ETL tool?
Not on its own. Airflow is an orchestrator: it schedules, retries and monitors tasks, which may call an ingestion tool, run SQL or start dbt. Many teams use Airflow together with Airbyte or Fivetran. See Apache Airflow vs Airbyte.
Is Talend still free?
No. Qlik retired Talend Open Studio on 31 January 2024. Talend is now sold as Qlik Talend Cloud and the commercial Talend Studio, both with free trials.
Did Fivetran buy dbt?
Fivetran and dbt Labs merged in an all-stock deal announced on 13 October 2025 and completed on 1 June 2026. The combined company operates as Fivetran + dbt Labs, and dbt Core remains open source under Apache 2.0.
Do I need an ETL tool if I only use one database?
Often not. If reporting runs on the same database, scheduled SQL (for example INSERT ... SELECT into reporting tables) may be enough. ETL tools earn their place when data comes from several systems or a separate warehouse. See data warehouse vs database.
Sources
- Fivetran pricing
- Fivetran usage-based pricing documentation
- Fivetran 2025 pricing updates FAQ
- Fivetran deployment models
- Fivetran and dbt Labs complete merger (press release)
- Fivetran agreement to acquire Census (press release)
- Hevo Data pricing
- Airbyte pricing
- Airbyte licences
- Meltano repository (MIT)
- dlt by dltHub
- Informatica Cloud Data Integration
- Salesforce completes acquisition of Informatica
- Qlik Talend Cloud pricing
- Talend Open Studio retirement (Qlik)
- Stitch (now part of Qlik)
- Matillion pricing
- Apache Airflow announcements
- Apache Airflow 3.0 release post
- Amazon MWAA supported Airflow versions
- Google Managed Service for Apache Airflow (Cloud Composer) versions
- Astronomer pricing
- Dagster pricing
- Prefect pricing
- Prefect repository (Apache 2.0)
- AWS Glue pricing
- What is Data Factory in Microsoft Fabric
- Fabric Data Factory vs Azure Data Factory
- Azure Data Factory pricing
- Google Cloud Dataflow pricing
- Google Cloud Datastream overview
- dbt pricing
Checked October 2026.
How we research these guides: our editorial method.