Time series databases store measurements stamped with a time, from server metrics to factory sensors and trade ticks.
This guide ranks thirteen of them on the engineering hours of the opening quarter, the invoice after retention and failover are added, the route out for your data, and whether it runs without a vendor.
Vendors can pay for more visibility on this website, for example a higher place in a ranking or more room on a page. Commercial relationships can also affect which products we feature and how prominently. Every vendor link is marked the same, whatever the arrangement behind it. How that works.
In short
What time series databases software does
A time series database stores and queries values indexed by timestamp, handling high write rates, retention policies and time-window aggregation for monitoring, industrial and financial measurements.
Every time series databases tool was judged on five points in a fixed order: setup effort, cost in practice, how data is exported, how easily you can leave, and the buyer it was designed for. Challengers hold places 1 to 3 wherever the category has them. Established brands begin at place 4. Our how we work page explains the choice of these five and why we give no mark out of ten.
Time series vendor that prints the rates of its cloud meters
Ranked #1 of 13 in Best Time Series Databases in 2026.
Free tierPublished pricingNorth America
Start on the Serverless free tier and watch which of the four meters grows fastest. Thirty days of retention on that tier means history is not a reason to stay on it, so plan the paid step early.
Enterprise adds high availability for self-managed clusters on a 30-day trial. Telegraf Enterprise is a separate line on the pricing page, so an agent fleet adds cost beyond the database itself.
What stands out
Metered cloud
Free tier
Self-managed Enterprise
Where it costs you
Four separate meters on Serverless make the monthly bill hard to forecast before real traffic arrives.
Core carries no high availability, so a resilient setup means a custom-priced edition.
The home page names no licence for Core, so reuse rights are unconfirmed.
Right for
Small teams wanting a cloud database with a visible, itemised price list
Wrong for
Buyers who need high availability without speaking to sales
United StatesUsage-based cloud with a free tier; Dedicated and Enterprise custom-priced
Hourly-billed time series service from the creators of TimescaleDB
Ranked #2 of 13 in Best Time Series Databases in 2026.
Pricing on requestNorth America
The plan ladder is the useful part. Performance covers up to four services; Scale removes that cap, adds a SOC 2 report and recovery of up to 14 days.
Sizing is chosen per service, and storage is billed on what is actually held, so an over-provisioned test cluster costs real money while it runs. Support at a one-business-day target comes with every plan, and faster response needs Enterprise.
What stands out
Hourly billing
30-day trial
TimescaleDB creators
Where it costs you
After the 30-day trial there is no continuing free tier, so a hobby project pays from day 31.
HIPAA compliance and SAML single sign-on sit only on the custom-priced Enterprise plan.
Performance caps a service at 8 CPU and 32 GB of memory, which forces a plan change as data grows.
Right for
Teams wanting predictable hourly billing and a clear plan ladder
Wrong for
Projects needing a permanent free instance after testing ends
United StatesHourly compute and storage billing on two plans; Enterprise custom; 30-day trial
Open source metrics database with paid Enterprise and managed Cloud tiers
Ranked #3 of 13 in Best Time Series Databases in 2026.
Open sourcePricing on requestElsewhere
Treat the Prometheus claim as a customer's experience, not a guarantee: it is a quote, and your own queries decide.
The practical path is to run the open source build beside your current setup, then price Enterprise or Cloud only if support or managed hosting matters. Enterprise users of the logs product get twelve months of backported fixes on its first long-term-support release.
What stands out
Prometheus replacement
Open source core
Managed Cloud
Where it costs you
Enterprise and Cloud carry no prices on the landing page, so budgeting starts with a conversation.
No open source licence is named on the pages read, which a legal review has to settle.
The vendor's registered location is not given on the site.
Right for
Teams replacing a Prometheus setup who want paid support available
Wrong for
Buyers who must see prices and the legal entity before talking
Not stated by the vendorOpen source core; Enterprise, Cloud and Anomaly Detection sold separately, no prices shown
Apache-licensed time series database with a monthly managed plan
Ranked #4 of 13 in Best Time Series Databases in 2026.
Open sourcePricing on requestNorth America
Storage-based billing suits workloads where retention is long and ingest is modest, since the plan advertises unlimited retention and charges for what is stored. The self-run core lets you test the engine before talking to anyone.
What you cannot test that way are the scaling and availability additions, which Enterprise carries. Ask for the proof of concept early, because the contact route is the only way to price the bring-your-own-cloud option.
What stands out
Apache-2.0 core
Managed monthly
Bring your own cloud
Where it costs you
The pricing page offers a live demo but no trial for either Enterprise plan.
Bring-your-own-cloud pricing starts with a request for a proof of concept.
Scaling, availability and security features belong to Enterprise, not the Apache-2.0 core.
Right for
Teams wanting an Apache-licensed core with a managed route later
Wrong for
Buyers needing a trial before committing to any paid plan
United StatesOpen source core; managed Enterprise billed monthly; bring-your-own-cloud custom
Community monitoring system whose time series store is local to each server
Ranked #5 of 13 in Best Time Series Databases in 2026.
Free tierOpen sourcePublished pricingElsewhere
Nobody can raise the price or retire a plan, which is the strongest independence story in this list. The cost is operational: you run the servers, size the disks and wire up Alertmanager.
Because each server keeps its own data on local storage, a growing estate pushes you toward a long-term store, and Grafana Mimir and VictoriaMetrics both appear here for that reason. Expect to learn PromQL before dashboards mean anything.
What stands out
Apache 2 licence
PromQL queries
CNCF graduated
Where it costs you
Local storage on each server means long retention and multi-server views need extra components.
Notifications and silencing depend on a separate Alertmanager that you also have to run.
The site names no vendor and no hosted option, so support is community only.
Right for
Teams monitoring services who accept running the whole stack themselves
Wrong for
Teams wanting a single hosted service with support attached
Not stated by the vendorFree to run under the Apache 2 licence; no vendor plans
Apache-licensed database sold as hourly cloud plans or run on-prem
Ranked #6 of 13 in Best Time Series Databases in 2026.
Self-hostablePublished pricingEurope
For a time series buyer the useful split is Dedicated against Custom. Dedicated is billed hourly, adds uptime SLAs and private links to AWS and Azure, and stops at nine nodes. Anything larger means a conversation.
Invoice and marketplace payment sit alongside card payment. Under the Apache 2.0 core, a design can be tried on your own hardware before choosing between the managed STACKIT cluster and a self-run setup.
What stands out
Hourly cloud
Apache 2.0 core
Run on-prem or edge
Where it costs you
The entry plan's limits are given twice on one page and do not match.
Four legal entities are listed, and the imprint does not name the operator.
Beyond nine nodes on Dedicated, price is only available through the Custom plan.
Right for
Buyers wanting hourly cloud billing with an on-prem or edge fallback
Wrong for
Teams that need a single named contracting entity
AustriaHourly cloud plans with a free entry plan; Custom plan on contact; Apache 2.0 core
AGPLv3 long-term store for Prometheus data, self-run or in Grafana Cloud
Ranked #7 of 13 in Best Time Series Databases in 2026.
Free tierSelf-hostablePublished pricingElsewhere
Three routes exist and their costs differ in kind. Self-run Mimir costs engineering time. The Grafana Cloud free plan allows 10,000 active series and 14 days of retention; Pro adds a monthly platform fee plus a per-1,000-series rate and 13 months retention.
Enterprise needs a yearly minimum commitment. Since the pricing page never says Mimir, confirm in writing that the metrics plan is the product you mean.
What stands out
AGPLv3 licence
Prometheus long-term store
Multi-tenant
Where it costs you
AGPLv3 can restrict how a company embeds or redistributes it inside its own product.
The cloud pricing page does not name Mimir, so the price link is an inference.
It stores Prometheus data and does not replace the Prometheus scrapers feeding it.
Right for
Teams already running Prometheus who have outgrown its local storage
Wrong for
Buyers wanting a standalone database rather than an add-on store
Not stated by the vendorAGPLv3 self-hosted; Grafana Cloud metrics tiers from a free plan to a yearly commitment
Columnar analytics engine used for time-stamped events, open source or hosted
Ranked #8 of 13 in Best Time Series Databases in 2026.
Open sourcePricing on requestNorth America
Here ClickHouse competes as a general analytics engine, not a purpose-built time series store. The upside is range: the same Apache 2.0 server runs locally, so a trial on a month of real data costs only effort.
The Basic tier caps storage at 1 TB in one availability zone, and Scale removes that cap. Retention and downsampling must be built in your own schemas, which is the trade for that flexibility.
What stands out
Open source server
Usage-based cloud
Trial credit
Where it costs you
Backups on Basic are kept for a day, which is short for data you cannot rebuild.
Rates for Basic and Scale are behind a calculator, so price comparison takes an extra step.
Table design and retention rules are left to the buyer.
Right for
Buyers storing metrics next to logs and events in one engine
Wrong for
Teams wanting ready-made downsampling without designing tables
United StatesUsage-based cloud tiers with 30-day trial credit; open source server; BYOC by quote
Industrial data platform priced by tag count, self-hosted or managed cloud
Ranked #9 of 13 in Best Time Series Databases in 2026.
Self-hostablePublished pricingElsewhere
Count your tags before anything else, because the whole price list hangs on that number. The self-hosted entry tier needs no payment but excludes standard support, and the 24x7 premium option has no listed price.
A perpetual licence costs roughly two and a half times the annual one at the same tag count, so the decision turns on how many years you expect to run it. Industrial protocol connectors come built in.
What stands out
Priced by tags
OPC, MQTT, Kafka
Self-hosted licence
Where it costs you
Every step up in tag count moves you into a new price bracket.
High availability adds thirty per cent to the licence cost.
The legal page names no entity, address or country.
Right for
Factory and sensor projects with a known, countable set of tags
Wrong for
Teams whose data volume is hard to express as tags
Not stated by the vendorAnnual or perpetual licence by tag count; monthly cloud packages; free use up to 5,000 tags
Hadoop and HBase time series store maintained by a community
Ranked #10 of 13 in Best Time Series Databases in 2026.
Pricing on requestElsewhere
The pitch is keeping full granularity forever, which suits teams that refuse to downsample. The price is the substrate: HBase must be sized, patched and backed up before OpenTSDB stores a single point.
Nothing on the site shows a company behind it or a paid tier, so the support path is asking the community. Pick it when an HBase estate already exists, and skip it when starting from nothing.
What stands out
Runs on HBase
Millisecond writes
Raw data kept
Where it costs you
It needs a Hadoop and HBase cluster underneath, which is a large operating commitment.
No licence is stated on the site, so usage rights need checking in the repository.
Community channels are the only help route the site names.
Right for
Organisations that already operate HBase and want full-granularity raw storage kept
Wrong for
Anyone lacking Hadoop skills or needing vendor support
Not stated by the vendorNo vendor pricing; the site names no licence or commercial offering
Apache project for time series from connected devices, edge to cloud
Ranked #11 of 13 in Best Time Series Databases in 2026.
Open sourcePricing on requestNorth America
Design intent is clear from the site: collect at the edge, synchronise to the cloud, analyse alongside Spark or Flink. Time alignment across devices and sensors is a stated query function.
What the pages never show is how to buy help, so budget internal engineering time. One-click cloud installation and a desktop tool lower the first-day effort, but a production sizing plan still has to come from you.
What stands out
IoT time series
Edge-cloud design
Flink and Spark links
Where it costs you
Neither a paid plan nor a support contract is mentioned on the site.
The pricing address returns a not-found page.
Licence terms come from the foundation page, not from the project site itself.
Right for
Device fleets needing edge and cloud storage under one project
Wrong for
Buyers wanting commercial support and a hosted plan
United StatesOpen source Apache project; the site lists no plans or prices
Real-time analytics database for streaming and batch data under Apache
Ranked #12 of 13 in Best Time Series Databases in 2026.
Open sourcePricing on requestNorth America
Choose Druid for slicing fresh events, not for storing metrics. Query-on-arrival lets dashboards read data as it lands, and tiering puts recent data on faster storage. Schema auto-discovery shortens first loads.
The gap is operational: the site lists no hosted offering and no price, so you carry deployment on a loosely coupled architecture with a deep storage layer that you also have to provide.
What stands out
Stream ingestion
Query tiering
Schema auto-discovery
Where it costs you
It describes itself as an analytics database, so time series features are not its focus.
The site gives no deployment or hosting options.
Pricing and paid support are both absent from the site text read.
Right for
Event analytics teams needing fast queries over fresh streaming data
Wrong for
Teams wanting a metrics store with built-in alerting rules
United StatesOpen source Apache project; no paid plan on the pages read
Market-data time series engine from KX with opaque usage pricing
Ranked #13 of 13 in Best Time Series Databases in 2026.
Pricing on requestElsewhere
The marketing has moved towards AI features, yet the buying question is simpler: what does a month of your memory footprint cost? The slider suggests billing follows memory, so a large in-memory working set raises the bill directly.
Market Data and Surveillance offerings are custom priced. Get the figures in writing, plus the trial terms, before committing, and ask whether GPU acceleration is billed on top of the memory slider.
What stands out
Memory-sized pricing
Vector search
Market data
Where it costs you
Pricing arrives only through a sales conversation, whatever the slider suggests.
A trial link appears without describing its limits or duration.
The legal page names two companies without an address or country.
Right for
Trading and surveillance teams with market data already in hand
Wrong for
Anyone needing price clarity or a modest first project
Not stated by the vendorUsage-based monthly by memory size, no prices shown; market data offerings quoted
A time series database stores and queries values indexed by timestamp, handling high write rates, retention policies and time-window aggregation for monitoring, industrial and financial measurements. Differences that change the outcome seldom sit in the features. Here is the sequence we would follow.
01
Work out how much a sales call is worth to you
8 vendors on this page sell by quote, while 5 publish a pricing model you can read today: InfluxDB, Prometheus, CrateDB, Grafana Mimir, TDengine.
02
Ask which country the contract runs through
Headquarters range over 3 countries and 3 regions: North America 6, Elsewhere 6, Europe 1. Legal exposure follows the vendor, not only the data centre, so read the contracting entity before you sign.
03
Check the open source options
Source code is available for VictoriaMetrics, GreptimeDB, Prometheus, ClickHouse, Apache IoTDB, Apache Druid. Running it yourself swaps the licence fee for your own upkeep.
Retention and series count set the bill, not just ingest rate
The unit each vendor meters decides what grows your invoice. Grafana Mimir through Grafana Cloud charges per thousand active series, with 14 days of retention on the free plan and 13 months on Pro. InfluxDB meters megabytes written, query executions, stored gigabyte-hours and data sent out, so a chatty dashboard costs money. TDengine counts tags, which is easy to predict for a factory and awkward for a service fleet.
Tiger Data bills configured compute plus the storage actually held, so an idle but large cluster still costs. Before comparing quotes, convert your own estate into each unit: active series, tags, megabytes per day, and gigabytes retained, then compare the totals rather than the headline rates. A plan that looks cheap per unit can still be the dearest once your own volumes are applied, so run the sums for a busy month as well as an average one.
Count active series, tags and daily megabytes from a real system before asking for prices.
Ask what retention each plan includes and what extending it costs.
Model query traffic separately where the vendor meters queries.
Monitoring metrics and device data are different markets
Prometheus, VictoriaMetrics, Grafana Mimir and OpenTSDB all grew out of server monitoring, where series are labelled by name and key-value pairs. TDengine lists OPC, MQTT and Kafka connectors, and Apache IoTDB targets millions of low-power devices, so those two start from the plant floor. kdb+ from KX points at market data and surveillance.
ClickHouse and Apache Druid describe themselves as analytics engines, so they suit event data but leave retention and downsampling design to you. Picking a monitoring store for sensor data, or the reverse, is the commonest wrong turn: the protocols, the cardinality and the people who maintain it all differ. Ask each shortlisted vendor for a reference customer whose data looks like yours, and run your own sample through a trial before signing.
Write down your data source: servers, devices or market feeds, before shortlisting.
Check the protocols the product ingests natively, not through a bridge.
Ask whether the vendor calls it a time series database or an analytics database.
High availability is where free tiers end
Several vendors keep resilience on the paid side. InfluxDB 3 Core has no high availability, while Enterprise and Dedicated are priced by contact. TDengine charges thirty per cent extra on its licence plans for high availability. Tiger Data puts multi-node high-availability replicas on the Scale plan and recovery of 14 days against 3 on Performance.
Mimir replicates for availability, but you operate that yourself unless you buy Grafana Cloud. Prometheus keeps data on local disks, so a lost disk is a lost history. A cheap first quote that excludes failover is not the real price, because production nearly always needs it. Put the failover question in writing, and ask what happens to the invoice when a second node or replica is switched on.
Ask which plan introduces replicas and failover, and the surcharge for them.
Find out the recovery window in days, and who restores it.
Test a node failure during the trial rather than trusting the data sheet.
Who maintains it when no company sells it
Prometheus is run by a foundation project, OpenTSDB credits only its authors, and Apache IoTDB and Apache Druid sit under the Apache Software Foundation with no plan or price on their sites. That is independence in its purest form, and also silence when something breaks: OpenTSDB leaves you with community channels alone. Vendor-backed alternatives fill the gap differently.
ClickHouse, GreptimeDB and CrateDB ship Apache-licensed cores with paid hosting, and Mimir is AGPLv3, a licence with obligations for embedded use. Decide whether you can staff the operations yourself, and read the licence in the repository rather than the marketing page. Check too whether a cloud plan can export data into a self-run copy of the same engine, because that is the exit route you will want if prices change.
Name the engineer who will own upgrades and restores before choosing a community project.
Read the licence file in the repository, especially for AGPLv3 software.
Confirm a paid support route exists if your budget needs a vendor to call.
What goes wrong most often when buying time series databases software
Choosing on write speed in a benchmark while ignoring how many series or tags the real estate produces.
Treating a monitoring store such as Prometheus as the home for years of sensor history.
Assuming the free or open source edition includes failover, then discovering it is an Enterprise feature.
Signing a perpetual or annual licence before counting tags, devices or active series for the following year.
07
Frequently asked questions
8 answers
What is the best time series databases in 2026?
InfluxDB leads our ranking of 13. InfluxData Inc. of San Francisco prints its Serverless meter: megabytes written, query executions, gigabyte-hours stored and gigabytes sent out.
The free tier keeps data for 30 days with capped ingestion. The Core edition has no high availability and community support only, so production buyers end up on Enterprise or Dedicated, both priced by contact.
Which time series databases tools publish their pricing?
5 of the 13, with the pricing model each one publishes:
InfluxDB: Usage-based cloud with a free tier; Dedicated and Enterprise custom-priced.
Prometheus: Free to run under the Apache 2 licence; no vendor plans.
CrateDB: Hourly cloud plans with a free entry plan; Custom plan on contact; Apache 2.0 core.
Grafana Mimir: AGPLv3 self-hosted; Grafana Cloud metrics tiers from a free plan to a yearly commitment.
TDengine: Annual or perpetual licence by tag count; monthly cloud packages; free use up to 5,000 tags.
The other 8 quote per organisation.
Is there a free time series databases tool?
InfluxDB, Prometheus, Grafana Mimir come with a no-cost entry point, a free plan or a self-hosted edition.
Where are these time series databases vendors established?
Vendor countries on this page, North America 6, Elsewhere 6, Europe 1:
InfluxDB: United States.
Tiger Data: United States.
VictoriaMetrics: Not stated by the vendor.
GreptimeDB: United States.
Prometheus: Not stated by the vendor.
CrateDB: Austria.
Grafana Mimir: Not stated by the vendor.
ClickHouse: United States.
TDengine: Not stated by the vendor.
OpenTSDB: Not stated by the vendor.
Apache IoTDB: United States.
Apache Druid: United States.
kdb+: Not stated by the vendor.
Which time series databases tools are open source?
Which time series databases tools can you host yourself?
CrateDB, Grafana Mimir, TDengine. The remaining 10 run only on the vendor's servers.
What should you use instead of InfluxDB?
Tiger Data and VictoriaMetrics are the next two on this page. Tiger Data is for teams wanting predictable hourly billing and a clear plan ladder; VictoriaMetrics is for teams replacing a Prometheus setup who want paid support available.
Who should not buy InfluxDB?
Buyers who need high availability without speaking to sales. Four separate meters on Serverless make the monthly bill hard to forecast before real traffic arrives..
Sell time series databases and missing from the ranking? Send us a line on what the product does and who buys it. Whether it is added is our decision, though commercial arrangements can also affect it, and the terms of a listing sit on the software advice page.
