Legacy SIEM vs modern SIEM: what changed and how to choose
Legacy SIEM vs modern SIEM: where logs live, how detections are written, what happens after an alert fires, and what each costs, with real product examples.
The short answer
Choose a modern SIEM if you're buying your first SIEM, your log volume is growing faster than your budget, or analyst time is the real limit; stay on a legacy SIEM if it already covers your sources, your team has years of tuned detections in its query language, and your contract makes a move expensive. The difference is design, not brand: a legacy SIEM ingests everything into its own storage, bills by volume, keeps detections in a proprietary query language, and ends at an alert queue, while a modern SIEM searches data where it lives, writes detections as code, and has agents investigate every alert. Most incumbents now sell pieces of both, so check each design choice rather than the label.
Which to choose
Choose Legacy SIEM if
- Your current SIEM covers the sources you need and the bill is predictable.
- Your team has years of tuned SPL or KQL detections, dashboards, and playbooks, and the people who maintain them.
- You must run on-premises, which Splunk Enterprise Security supports and cloud-only SIEMs don't.
- Most of your logs come from Microsoft 365, Azure, and Defender, where Sentinel ingests several sources free.
- Your renewal is years away, and you'd rather add agents to the SIEM you have than migrate mid-contract.
Choose Modern SIEM if
- You're choosing your first SIEM, or a renewal is coming and you're open to moving.
- You leave log sources out today because ingesting them costs too much.
- You want logs in your own cloud storage, in open formats other tools can read.
- Your detection engineers would rather write Python with tests and version control than a vendor query language.
- Analyst hours, not storage, are your limit, and you want every alert investigated rather than the ones someone reaches.
The two designs at a glance
| Legacy SIEM | Modern SIEM | |
|---|---|---|
| Where logs live | The vendor's storage: Splunk indexes, a Sentinel workspace, Google's cloud | Your storage: Cotool in your cloud bucket, Panther in your Snowflake or Databricks |
| Detecting on data you didn't ingest | Through add-ons, such as Splunk federated search for Amazon S3 | Built in; Cotool queries and detects on sources it doesn't store |
| Detection language | SPL (Splunk), KQL (Sentinel), YARA-L (Google), CQL (CrowdStrike) | Python (Panther, Cotool) |
| After an alert fires | An analyst queue, with automation built separately (Splunk SOAR, Sentinel playbooks on Logic Apps) | Agents investigate every alert; in Cotool each disposition tunes the detection that fired |
| Where AI fits | Assistants and MCP servers added to the platform, such as the Sentinel and Google SecOps MCP servers | Agents do the first pass on detections and triage |
| Long retention | Cheaper tiers or archives (Sentinel data lake up to 12 years; Splunk archive restored before search) | Logs stay in your bucket in open formats such as Parquet and are searched in place |
- Where logs live
- Legacy SIEMThe vendor's storage: Splunk indexes, a Sentinel workspace, Google's cloudModern SIEMYour storage: Cotool in your cloud bucket, Panther in your Snowflake or Databricks
- Detecting on data you didn't ingest
- Legacy SIEMThrough add-ons, such as Splunk federated search for Amazon S3Modern SIEMBuilt in; Cotool queries and detects on sources it doesn't store
- Detection language
- Legacy SIEMSPL (Splunk), KQL (Sentinel), YARA-L (Google), CQL (CrowdStrike)Modern SIEMPython (Panther, Cotool)
- After an alert fires
- Legacy SIEMAn analyst queue, with automation built separately (Splunk SOAR, Sentinel playbooks on Logic Apps)Modern SIEMAgents investigate every alert; in Cotool each disposition tunes the detection that fired
- Where AI fits
- Legacy SIEMAssistants and MCP servers added to the platform, such as the Sentinel and Google SecOps MCP serversModern SIEMAgents do the first pass on detections and triage
- Long retention
- Legacy SIEMCheaper tiers or archives (Sentinel data lake up to 12 years; Splunk archive restored before search)Modern SIEMLogs stay in your bucket in open formats such as Parquet and are searched in place
Typical designs, with examples from vendor documentation as of November 2026. Many products now mix both, so check each row for the product you're evaluating.
Pricing at a glance
| Legacy SIEM | Modern SIEM | |
|---|---|---|
| Typical billing unit | GB ingested or retained, or compute (Splunk workload pricing) | Varies by vendor; ask whether volume, compute, or seats drive the bill |
| Published prices | Sentinel and Elastic publish per-GB prices; Splunk and Google sell by quote | Not published for Cotool or Panther |
| Storage you pay for directly | Included in the SIEM bill | Your cloud provider bills it, as with Amazon Security Lake on S3 |
- Typical billing unit
- Legacy SIEMGB ingested or retained, or compute (Splunk workload pricing)Modern SIEMVaries by vendor; ask whether volume, compute, or seats drive the bill
- Published prices
- Legacy SIEMSentinel and Elastic publish per-GB prices; Splunk and Google sell by quoteModern SIEMNot published for Cotool or Panther
- Storage you pay for directly
- Legacy SIEMIncluded in the SIEM billModern SIEMYour cloud provider bills it, as with Amazon Security Lake on S3
Published pricing as of November 2026.
What each approach is
A legacy SIEM is built around one idea: collect every log into the SIEM's own storage, index it, and then search and correlate. Detections are scheduled searches in the vendor's query language, matches become alerts, and alerts go to a queue that analysts work through. Its cost is built in: a source has to be ingested and paid for before anything can detect on it, and every alert waits for a person. Splunk Enterprise Security follows this design: it runs on a Splunk platform licensed by daily ingest or by compute, and buying Enterprise Security adds no ingest capacity on top of that license. Microsoft Sentinel runs detections as KQL analytics rules on data in a Log Analytics workspace, and Google Security Operations writes them in YARA-L over data in Google's cloud.
A modern SIEM changes three things. Storage moves out of the SIEM: logs sit in your own cloud object storage in open formats, or stay where they already are and get searched in place. Detections become code in a general-purpose language, with tests and version control. And agents take the first pass at the work analysts used to do by hand, from writing and tuning detections to investigating each alert. Cotool is built this way: logs stay in your bucket as Parquet, detections are Python that agents write, test, and tune, and an agent investigates every alert. Panther, now part of Databricks and being folded into its Lakewatch SIEM, also writes detections in Python and runs them against data in a lakehouse.
The line runs through products more than between vendors. Sentinel has added a data lake tier in an open format, notebooks, and an MCP server for agents, though analytics rules have limits on data that sits only in the lake. Splunk sells federated search for Amazon S3 as an optional add-on, with a 10 TB data volume limit by default. CrowdStrike Falcon Next-Gen SIEM offers federated search into S3 and Snowflake, and Google Security Operations has an MCP server too. So rather than sorting vendors into two camps, check each product against the differences below.
Key differences
Where logs live, and who owns them. In a legacy SIEM the vendor stores your logs, and the bill grows with every source you add. Leaving means exporting years of data or giving up access to it. A modern SIEM keeps logs in storage you control: Cotool searches files in your own bucket in place, and Panther runs on your Snowflake or Databricks. Other tools can read the same data, and leaving at renewal doesn't mean exporting your history.
What gets left out. When every gigabyte costs SIEM rates, teams decide which sources they can afford, and the ones they skip become blind spots. Cotool's answer is storage cheap enough to onboard every source you need, plus federated queries that search and detect on sources it doesn't store, including an existing Splunk. Source health checks raise an alert when a source goes quiet, so a broken pipeline doesn't become a silent gap.
How detections are written. Splunk detections are scheduled searches in SPL, Sentinel's are KQL analytics rules, Google's are YARA-L rules, and CrowdStrike's searches use CQL. Each language works only inside its vendor's product, so years of content become a rewrite when you move. Modern platforms write detections as code in Python: the Databricks announcement describes Panther detections that teams write, unit-test, version-control, and deploy through CI/CD. In Cotool, agents write, test, and tune detections in Python, every new detection and every tune is backtested against your history before it deploys, and a library of hundreds of detections mapped to MITRE ATT&CK gives you coverage on day one.
What happens after an alert fires. The legacy answer to alert volume is to make the queue smaller: Splunk's risk-based alerting groups findings by user or system before anyone is paged, Splunk SOAR is part of Enterprise Security Premier, and Sentinel automates with playbooks on Azure Logic Apps, which you build and maintain. Splunk's product page also describes a triaging agent in Enterprise Security that "evaluates, prioritizes, and explains alerts" so analysts can focus on the most critical ones; the analyst still works the queue. A modern SIEM puts an agent on every alert.
Take an MFA push-fatigue alert from Okta. In a queue-based SIEM it waits until an analyst pulls the user's sign-in history, checks the EDR for activity on their laptop, and closes or escalates it by hand; a playbook can automate parts of that only if someone has built and maintains it. In Cotool, the triage agent picks up the alert when it lands, gathers that context through its integrations, and closes or escalates it with the evidence attached in built-in case management. A benign disposition tunes the Python detection, and the tune is backtested against your history before it deploys, so a noisy rule gets quieter instead of piling up in the queue. State-changing steps, such as suspending the user, can require approval in the app or Slack.
Where AI fits. Incumbents add AI on top of the existing design: assistants inside the console, and MCP servers such as Sentinel's and Google's that let outside agents query the platform. The queue and the query language stay the same. A modern SIEM is built for agents to operate it. Cotool exposes the same logs over an API and MCP, so Claude Code or your own agents can search logs and run detections, and analysts search the full history without separate tools.
What it costs
Legacy SIEMs mostly bill by data volume. Sentinel's analytics tier is $4.30 per GB pay-as-you-go in East US, with commitment tiers that Microsoft says save up to 52%; several Microsoft sources ingest free, and everything else bills at those rates; the Microsoft Sentinel pricing guide walks through every meter. Elastic's serverless security tier starts at $0.09 per GB ingested and $0.017 per GB-month retained. Sumo Logic moves the meter rather than removing it: its Flex plan lists $0 ingest and bills by the volume each search scans, including the queries behind dashboards and monitors. Splunk prices by GB ingested per day or by compute, and Google Security Operations sells ingestion-based packages; neither publishes prices. Long retention moves to cheaper tiers: Sentinel's data lake tier holds data for up to 12 years and bills storage on compressed size at a 6:1 ratio, while Splunk's archive keeps data for up to 10 years in total but has to restore it before you can search it, typically within 24 hours.
Modern SIEMs move the storage line to your cloud bill. Amazon Security Lake, a storage layer rather than a SIEM, shows the shape: AWS charges per GB ingested, $0.75 for CloudTrail and $0.25 for other AWS logs in US East, plus S3 storage, and it converts logs from supported AWS services to the OCSF schema in Apache Parquet. Cotool and Panther don't publish prices, so you can't compare them from a price list the way you can Sentinel or Elastic. Ask any vendor what drives the bill (volume, compute, or seats), price it against your daily volume, and count analyst hours as part of the cost: a SIEM that costs less per GB but leaves every alert to a person may not cost less overall.
How Cotool approaches it
Cotool's AI-native SIEM follows the modern column of the table in every row. Logs go into your own cloud object storage in open formats such as Parquet, and Cotool searches them in place, so they stay yours and never need exporting. Federated search reaches sources Cotool doesn't store. Custom and in-house sources are parsed automatically and validated on real events before they go live. Detections are Python in notebooks, written, tested, and tuned by agents and backtested before every deploy. Every alert is investigated, with each disposition feeding back into the detection that fired, and the same agents handle alerts from your other tools through Cotool's AI SOC.
There are three ways in. Teams without a SIEM connect sources and start from the detection library. Teams replacing one run Cotool alongside it, bring their existing rules, and move sources over at their own pace. If you're weighing specific replacements for Splunk, the Splunk alternatives guide compares the main options. Teams keeping their SIEM leave sources and detections where they are, move only what's too expensive to store, and use Cotool's agents for triage. The trade-off is that Cotool doesn't publish pricing, so sizing it starts with a demo rather than a calculator.
FAQ
- What is the difference between a legacy SIEM and a modern SIEM?
- A legacy SIEM collects every log into the vendor's storage before it can detect on it, bills by data volume, keeps detections in a proprietary query language such as SPL or KQL, and sends alerts to a queue for analysts. A modern SIEM keeps logs in your own storage or searches them where they live, writes detections as code in a general language such as Python, and has AI agents investigate alerts before a person sees them.
- Is Microsoft Sentinel a legacy SIEM?
- Microsoft Sentinel mixes both designs. Its analytics tier works the traditional way: logs are ingested into a Log Analytics workspace, billed per GB, and detected on with KQL analytics rules. Microsoft has added a data lake tier with long, cheap retention, notebooks for analysis, and an MCP server for AI agents, which are newer patterns built alongside that core.
- Do I have to replace my SIEM to get the benefits of a modern one?
- You don't have to replace a SIEM to get agent triage or cheaper storage for some sources. Several modern platforms run alongside an existing SIEM: Cotool, for example, can query an existing Splunk in place, keep the sources that are too expensive to store there in your own bucket, and have agents investigate the alerts either system raises.
- What is detection as code?
- Detection as code means writing detection rules as code in a general-purpose language, usually Python, and managing them like software: in version control, with unit tests, and deployed through CI/CD. Panther and Cotool both write detections in Python, while traditional SIEMs use their own query languages.
- Is a modern SIEM cheaper than a legacy SIEM?
- A modern SIEM is not automatically cheaper, and most modern vendors, including Cotool and Panther, don't publish prices. The savings usually come from storing logs in your own cloud object storage instead of paying a SIEM rate per GB, and from agents handling triage that would otherwise take analyst hours. Get quotes priced on your own daily volume and alert count.
Sources
- AI-Native SIEM Built for Agents - Cotool · accessed November 16, 2026
- AI SOC Without the Black Box - Cotool · accessed November 16, 2026
- Splunk pricing models · accessed November 16, 2026
- Licensing for Splunk Enterprise Security · accessed November 16, 2026
- Use detections to search for threats in Splunk Enterprise Security · accessed November 16, 2026
- Deployment considerations for Splunk Enterprise Security · accessed November 16, 2026
- Splunk Enterprise Security · accessed November 16, 2026
- Splunk Cloud Platform service details · accessed November 16, 2026
- Store expired Splunk Cloud Platform data in a Splunk-managed archive · accessed November 16, 2026
- Plan costs and understand pricing and billing - Microsoft Sentinel · accessed November 16, 2026
- Manage data tiers and retention in Microsoft Sentinel · accessed November 16, 2026
- Microsoft Sentinel data lake overview · accessed November 16, 2026
- What is Microsoft Sentinel's support for MCP? · accessed November 16, 2026
- Scheduled analytics rules in Microsoft Sentinel · accessed November 16, 2026
- Automation in Microsoft Sentinel · accessed November 16, 2026
- Azure Retail Prices API: Microsoft Sentinel, East US · accessed November 16, 2026
- Google SecOps packages overview · accessed November 16, 2026
- Single and multiple event rules in YARA-L · accessed November 16, 2026
- Configure MCP - Google Security Operations · accessed November 16, 2026
- Next-Gen SIEM - CrowdStrike · accessed November 16, 2026
- CrowdStrike Falcon Next-Gen SIEM Top 10 FAQs · accessed November 16, 2026
- Elastic Cloud Serverless pricing for Elastic Security · accessed November 16, 2026
- Panther - The Complete AI SOC Platform · accessed November 16, 2026
- Databricks Completes Acquisition of Panther: Accelerating the Security Lakehouse Era · accessed November 16, 2026
- Amazon Security Lake pricing · accessed November 16, 2026
- Sumo Logic pricing · accessed November 16, 2026





