The reason it’s so hard to compare data integration tools isn’t features. Every one of these will happily pull Salesforce, Postgres, and Stripe into your warehouse. The reason it’s hard is that they don’t agree on what they’re charging you for.
Fivetran bills by rows that changed. Airbyte bills by gigabytes moved, unless the source is an API, in which case it bills by rows again. Stitch bills by rows loaded into your warehouse. Estuary bills per-GB plus a flat fee per connector. Meltano charges you nothing and then bills you in engineering hours you forgot to count. Five tools, five units that don’t convert into each other. That’s the whole problem, and it’s why the same pipeline can cost $200 on one and $6,000 on another.
I’ve watched a mid-size team’s Fivetran bill go from “reasonable” to “who approved this” in one quarter because someone turned on history mode on a high-churn table. So let’s talk about the pricing models first, then features, because the pricing model is what actually decides this.
The five pricing units, side by side
Here’s the thing nobody tells you up front: the pricing unit determines which kind of data is expensive for you. Pick the wrong tool and you’re paying a premium on exactly the pattern your business generates most.
- Fivetran — Monthly Active Rows (MAR). A unique primary key that got inserted, updated, or deleted at least once this month. Counted once no matter how many times it changed.
- Airbyte — per-GB for database sources ($10/GB), per-million-rows for API sources ($15/M). Cloud entry is $10/month.
- Stitch — rows replicated per month, in fixed tiers. Standard is $100/month.
- Meltano — free and open source (Apache 2.0). You pay for the servers and the person who runs it.
- Estuary — $0.50/GB moved, plus $100/month per connector for the first six, $50/month each after.
Sit with that MAR-vs-GB difference for a second. If you sync a 500 GB table where only 2% of rows change monthly, Fivetran charges you for that 2% and Airbyte charges you for moving the whole thing on the initial sync. Flip it — a small table that updates constantly — and the math inverts. Neither is “cheaper.” They’re cheaper for different data.
Fivetran: the one everyone benchmarks against
Fivetran is the default because it just works. Hundreds of connectors, near-zero maintenance, and when a source API changes their schema at 2am, Fivetran usually absorbs it before you notice. For a lot of teams that reliability is worth real money.
The pricing is where it gets spicy. As of 2026, MAR tiers run roughly $2.50 per million for the first 5M, dropping to $2.00 (5–20M), $1.50 (20–100M), and $1.00 above 100M — and those rates depend on your plan tier (Standard, Enterprise, Business Critical). Two changes landed recently that matter:
First, since March 2025, MAR is calculated per connector, not pooled across your account. That quietly raised bills for anyone who was benefiting from volume discounts across many small connectors — each one now climbs the tier ladder on its own.
Second, in January 2026 Fivetran added a $5 base charge per connection generating between 1 and 1M MAR (Free plan excluded). And deletes now count toward MAR. If you’ve got a lot of low-volume connectors, that $5-a-pop floor adds up in a way the per-million rate never shows you.
The failure mode is predictable. Fivetran is excellent value until your row churn scales faster than your revenue. High-frequency updates, CDC on chatty tables, or history mode on a big dimension table — any of those and the bill goes vertical. Watch it monthly, not quarterly.
Airbyte: cheaper at volume, if you can tolerate the rough edges
Airbyte’s pitch is 600+ connectors, an open-source core you can self-host for free, and Cloud pricing that comes in meaningfully under Fivetran — often 50–70% less at equivalent volumes. The per-GB model ($10/GB database, $15/M rows API) rewards exactly the case Fivetran punishes: big tables that don’t change much relative to their size.
Two Airbytes exist and you should be clear on which you mean. Open-source Airbyte is free to run but it’s a real system to operate — you’re maintaining deployments, upgrades, and connector quirks yourself. Airbyte Cloud is managed and metered. There’s also a capacity/Data-Workers model on higher tiers where you pay for concurrent pipeline compute rather than raw volume, which changes the math again if you run many pipelines in parallel.
The honest trade-off: connector quality is more variable than Fivetran’s. The long tail of community connectors ranges from rock-solid to “works until it doesn’t.” For popular sources you’ll be fine. For an obscure SaaS with a weird API, budget time to test before you commit a production pipeline to it. Airbyte is where I’d start if cost is the driving concern and you have at least one engineer who won’t panic when a sync fails.
Stitch: the boring one, and that’s a compliment
Stitch is old, stable, and refreshingly simple to price. Standard is $100/month covering 5 million rows (some sources cite the tier up to a few hundred million depending on plan details), Advanced runs $1,500/month, Premium $3,000/month. Rows replicated, fixed tiers, no MAR gymnastics. You know your bill before you turn it on.
It’s built on the open-source Singer spec, so the connector philosophy overlaps with Meltano. Stitch was acquired by Talend in 2018, and Talend by Qlik in 2023 — which tells you where the roadmap energy is going, and it’s not into Stitch. Development has been quiet for years.
So Stitch fits a narrow but real slot: you want predictable pricing, your sources are mainstream, and you don’t need cutting-edge connectors or sub-hour latency. If that’s you, the simplicity is genuinely nice. If you need the newest SaaS connector or CDC streaming, look elsewhere — Stitch isn’t chasing those.
One note if you’re in healthcare: HIPAA BAA signing only comes on the Advanced and Premium tiers, and private-networking add-ons (PrivateLink, VPN, VPC peering) cost extra on top. The $100 plan is not the compliance plan.
Meltano: free like a puppy is free
Meltano Core is Apache 2.0 licensed and fully functional with zero license cost. It’s a CLI-first, code-defined, Singer-based orchestrator that data engineers who like version control and hate clicking around a SaaS UI tend to love. Your pipelines are config files. They live in git. You test them in CI. For a certain kind of team that’s the dream.
The cost that doesn’t show up on an invoice is operational. You’re running the infrastructure, managing the taps and targets, debugging Singer connector edge cases, and owning uptime. That’s fine if you have the engineering bandwidth and want full control — and genuinely cheaper than any SaaS at high volume. It’s a trap if you picked “free” to save money and then burned a senior engineer’s month keeping it alive.
There’s a managed Meltano Cloud offering priced in compute hours rather than rows (small teams often fit the entry tier), which softens the ops burden. But the core reason to choose Meltano is that you want pipelines as code and are willing to operate them. If you don’t want that, its main advantage isn’t for you.
Estuary: when latency is the point
Estuary is the CDC-first, streaming option in this group. Instead of batch syncs every few hours, it captures changes in near real-time. Pricing is $0.50/GB moved plus $100/month per connector for the first six ($50 each after) — and that per-GB rate is low enough that Estuary often undercuts Airbyte once you’re into tens of GB per month.
The reason to reach for it is latency. If you’re doing operational analytics, fraud signals, or anything where “the warehouse is 4 hours stale” is a real problem, batch tools like Stitch and standard Fivetran don’t solve it and Estuary does. It also does streaming transforms in-flight, which can save a downstream processing step.
The catch is the connector catalog is smaller than Fivetran’s or Airbyte’s, and streaming CDC is operationally more involved than batch — more to understand about your source database’s replication setup, WAL retention, and so on. If your actual requirement is “daily refresh is fine,” you’re paying complexity for latency you don’t need. But when you need it, the batch tools simply can’t compete.
A worked example: what does one real stack cost?
Let’s price a common mid-size setup: Salesforce + HubSpot + Stripe + a Postgres database on CDC. Say the Postgres table is 200 GB total with moderate monthly churn, and the SaaS sources generate a few million active rows a month combined. Rough, illustrative, 2026 list prices — your actual mileage depends heavily on churn and volume:
- Fivetran — the SaaS connectors are fine, but the Postgres CDC connector’s MAR is the wildcard. Moderate churn on a 200 GB table can land you anywhere from a few hundred to several thousand dollars a month. This is the setup where Fivetran either looks cheap or terrifying, entirely depending on that churn rate.
- Airbyte Cloud — the 200 GB database source dominates: roughly $2,000 on initial sync at $10/GB, then incremental after. The SaaS API sources add per-million-row charges. Predictable if your incremental syncs are small.
- Stitch — you’re on Advanced ($1,500/month) the moment you cross the Standard row cap, which a Postgres CDC source will do fast. Flat and predictable, but you’re paying for the tier ceiling, not your actual usage.
- Meltano — license cost $0. Real cost is the engineer-days to build and babysit four pipelines including a CDC tap. Cheapest at scale, most expensive in attention.
- Estuary — four connectors = $400/month in connector fees, plus $0.50/GB. For a CDC-heavy real-time need this is competitive and the latency is a genuine feature, not just a cost line.
The takeaway isn’t a winner. It’s that the same four sources swing by an order of magnitude depending on the tool, and the swing is driven almost entirely by how your Postgres table churns. Which is exactly why you have to know your data before you shop.
So which one
Rough decision guide, and I’ll actually commit to opinions:
- You want it to just work and cost is secondary → Fivetran. Set a billing alert and check it monthly.
- Cost matters and you have an engineer who won’t flinch → Airbyte, Cloud or self-hosted depending on how much ops you want to own.
- You want a predictable flat bill and your sources are mainstream → Stitch. Boring, stable, done.
- You want pipelines as code and have the team to run them → Meltano. Free is real if the hours are already yours.
- You need near-real-time, not daily batch → Estuary. Nothing else here competes on latency.
Before you sign anything, do one thing: run a two-week trial on your actual highest-volume source and read the metered bill. Every one of these tools is cheap in the demo and honest only on your real data. The connector you’re worried about is the one to test first.