Feature Stores for Machine Learning in 2026: Top 6 Compared
Fraud and risk teams that require query-time features · Pricing starts at: Quote only
Direct answer
Chalk leads the list, best for fraud and risk teams that require query-time features. Listed starting prices: SageMaker Feature Store from $0.45 per GB-month for standard online store storage, Vertex AI Feature Store from $0.08 per hour for a data processing node and Feast from $0 open source (Apache 2.0 licence). For a different need, Tecton covers large teams that require live serving and SageMaker Feature Store covers AWS teams that require managed serving. The remaining 3 options are covered below with prices and limits.
Choosing feature stores for machine learning usually comes down to price, fit and limits. This ProMapRanker guide sets 6 tools side by side on those three points, with the facts kept short and checkable.
Key takeaways
- The list opens with Chalk, aimed at fraud and risk teams that require query-time features.
- Databricks Feature Store suits teams already on Databricks lakehouse.
- Starting prices begin at $0.45 per GB-month for standard online store storage with SageMaker Feature Store.
- A fair warning about Chalk: no listed price and no free plan.
- Feast, Databricks Feature Store: these are the 2 tools out of 6 that offer a free plan.
Quick comparison
| Pick | Great for | Pricing starts at | Free tier | |
|---|---|---|---|---|
| Fraud and risk teams that require query-time features | Quote only | No | Visit | |
| Large teams that require live serving | Quote only | No | Visit | |
| AWS teams that require managed serving | $0.45 per GB-month for standard online store storage | No | Visit | |
| Google Cloud teams using BigQuery | $0.08 per hour for a data processing node | No | Visit | |
| Open source teams that self-host | $0 open source (Apache 2.0 licence) | Yes | Visit | |
| Teams already on Databricks lakehouse | $0.15 per DBU for Jobs Compute (Premium tier) | Yes | Visit |
Feature Stores for Machine Learning, one by one
Where Chalk fits: fraud and risk teams that require query-time features. Chalk is a live feature product for teams that require features computed at request time, instead of precomputed and stored in advance. For everyday use, the vendor lists online serving under 5 ms on its Velox-based engine, and it can run hosted, in your own cloud cluster, or in an air-gapped setup.
In daily use, ask for a sizing estimate against your own feature list. A marketplace listing displays $0.85 per credit, with credits including feature computation, live serving, and agent sandbox work, and your own cloud Pricing for Chalk is quote based, so budget talks start with the vendor. No free plan comes with Chalk.
Where it falls short: No listed price and no free plan.
- Features defined in plain Python and computed at request time
- Quick online serving on a Velox-based engine
- Resolvers are specific to Chalk and do not port to other keeps

Tecton works best for large teams that require live serving. Teams define features in Python, and Tecton deals with materialisation, online serving, and drift checks.
Points against: Now part of Databricks, so the standalone future is less clear.
Also, ask how Tecton features map to Databricks plans. Tecton does not show a fixed starting price, so you request a quote. Tecton has no permanent free option.
- Capable streaming and batch transformation support
- Managed live serving with monitoring
- No listed price and no free plan
Think of SageMaker Feature Store for AWS teams that require managed serving. Standard storage lists at $0.45 per GB-month, on-demand writes list at $1.25 per million write units, and on-demand reads list at $0.25 per million read units, with each write unit handling 1 KB and each read unit handling 4 KB.
In daily use, Provisioned capacity also bills for the capacity you reserve, even when you do not use it completely. On cost, SageMaker Feature Store starts like this: $0.45 per GB-month for standard online store storage. SageMaker Feature Store does not include a free plan.
Points against: In-memory store has a 5 GiB minimum charge per hour.
- In-memory online store for low latency reads
- Managed store inside SageMaker with on-demand and provisioned modes
- Writes each bill separately
Vertex AI Feature Store is designed for Google Cloud teams using BigQuery. The current version is built on BigQuery, which lets teams define features from BigQuery tables and serve them through Vertex AI.
Where it falls short: Legacy version closes on February 17, 2027.
In daily use, the legacy version will close completely on February 17, 2027, so new projects should start on the current version and existing users should plan a move. Price to start with Vertex AI Feature Store: $0.08 per hour for a data processing node. Vertex AI Feature Store does not include a free plan.
- Serving choices bundle optimised and Bigtable nodes
- BigQuery-native store in the current version
- Serving node costs add up when nodes run all day
Feast serves open source teams that self-host. The trade-off is operational work. It works with Redis for online serving and with warehouses for example BigQuery and Snowflake for offline storage.
Trade-offs to know: You run and monitor all infrastructure.
It supports on-demand transformations and streaming ingestion. On top of that, skip it when you need a managed service with a service level agreement. Feast pricing: $0 open source. A free plan is available with Feast.
- No licence fee under Apache 2.0
- Python-native feature definitions with point-in-time joins
- No managed hosting or support contract in the core project

Databricks Feature Store makes the most sense for teams already on Databricks lakehouse. Pricing is usage based in Databricks Units, known as DBUs, and the rate changes by workload.
Where it falls short: DBU rates vary by workload, level, cloud, region.
Worth adding: choose this store when your data and models already live on Databricks. Databricks Feature Store cost at the entry level: $0.15 per DBU for Jobs Compute. Databricks Feature Store has a free plan.
- Free Edition for learning and small work
- Tight fit with Unity Catalog, MLflow
- Cloud infrastructure is billed separately on classic compute
How we chose
The comparison looks at four things: starting price, free access, how well the feature set matches feature stores for machine learning, and the main limit of each tool. Seat minimums, billing terms and add-ons can move the real cost, so the starting prices here are a guide and not a quote. The order reflects how well each tool fits the most common buying scenario for feature stores for machine learning, then price clarity, then free plan access. It is a starting point, not a verdict.
Before you commit
- Ask Chalk and Tecton for a written quote, since no starting price is shown.
- Check what the free option really includes: Feast and Databricks Feature Store offer a free plan, and usage caps differ.
- Before choosing Chalk, weigh its main limit: no listed price and no free plan.
- Map your must-have features to a specific Chalk plan before you compare prices.
- Involve the people who will use Tecton daily; a short pilot beats a long feature checklist.
Frequently asked questions
Can I try these tools before paying?
Often, yes. Feast and Databricks Feature Store have a free plan. Use your own data and a real task during the test, and check what the free tier or trial limits before you rely on it.
Do these tools publish their prices?
Not all of them. Chalk and Tecton publish no starting price, so expect a quote request for those. The rest, SageMaker Feature Store included, show a starting price, though seats and billing terms still change the final cost.
Is Chalk the right pick for everyone comparing feature stores for machine learning?
No single tool suits every team. Chalk comes first because it fits fraud and risk teams that need query-time features, while Tecton suits large teams that need real-time serving and SageMaker Feature Store suits AWS teams that need managed serving. Match the tool to your own size, budget and workflow before deciding.
Which options for feature stores for machine learning have a free plan?
On this list, Feast and Databricks Feature Store include a free plan. Caps on seats, usage or features are common, so test the free plan against your real workload.
What is the best feature store for machine learning?
Chalk is the pick this page puts first, mainly for fraud and risk teams that require query-time features. If your situation differs, Tecton suits large teams that require real-time serving, and SageMaker Feature Store suits AWS teams that require managed serving. Compare price and limits before you decide.
Which feature store is best for live serving?
Chalk comes first here and is built for fraud and risk teams that require query-time features. Buyers with other needs may prefer Tecton (for large teams that require real-time serving) or SageMaker Feature Store (for AWS teams that require managed serving). Compare costs and limits first.
Sources
See where you really rank - block by block
ProMapRanker scans Google Maps across a grid of your service area. Simple monthly plans from $19, white-label on every plan.
Start free