Device model dimension
The device model dimension reports the specific device a visit came from. On the web it is inferred from the user-agent string or client hints; in apps the SDK reads it directly. This page explains the derivation, why web models are deliberately coarse for privacy, and how app data is more precise.
What this means
The device model dimension answers 'what hardware was this?' — an iPhone, a specific Android phone, a tablet model. On the web it is inferred: the analytics tool parses the user-agent string, or reads User-Agent Client Hints, and maps the tokens to a model.
In native apps the picture is cleaner: the GA4/Firebase SDK reads the model from the operating system directly, so app device-model data is generally more precise and reliable than web inference.
- Web: inferred from user-agent or client hints
- App: read directly by the SDK from the OS
- App model data is more precise than web inference
Why web models are coarse
Browsers have deliberately reduced the model detail in user-agent strings. Modern Safari and Chrome report generic identifiers rather than exact generations, because a precise model is a strong fingerprinting signal that can help re-identify a visitor across sites.
The result is that web device-model reporting is often a family ('Apple iPhone') rather than a generation ('iPhone 15 Pro'). Treat this as a privacy feature, not a data gap, and lean on app data when you genuinely need model-level precision.
- User-agent model detail has been frozen/reduced
- Precise models are strong fingerprinting signals
- Expect families on web, generations in apps
How it appears in analytics and logs
A device model value names the hardware behind a visit. Vague web values like 'Apple iPhone' without a generation are expected — browsers freeze model detail to limit fingerprinting.
Diagnostic use case
Identify which device models dominate your traffic so you can prioritize testing and fix model-specific rendering or performance issues.
What WebmasterID can help detect
WebmasterID derives device context from declared signals first-party and avoids fingerprinting, so model data stays coarse and privacy-safe by design.
Common mistakes
- Expecting exact web model generations that browsers no longer expose.
- Combining model with other fields in ways that fingerprint users.
- Comparing web model granularity to app model granularity directly.
Privacy and accuracy notes
Precise device models add fingerprinting entropy, which is why browsers reduce them. Avoid combining model with other high-entropy fields to re-identify users. This is educational, not legal advice.
Related pages
- Device brand dimension
The device brand dimension reports the manufacturer of the device behind a visit — Apple, Samsung, Google, and so on. It is the broader sibling of device model and is inferred from user-agent or client hints on the web, or read by the SDK in apps. This page explains derivation and where brand inference goes wrong.
- Device category: desktop, mobile, tablet
Device category groups visits into desktop, mobile, or tablet. It is derived from the user-agent string (increasingly, User-Agent Client Hints), so it is a classification, not a hardware fact. Tablets, desktop-mode mobile browsers, and foldables blur the boundaries, and the user agent can be spoofed.
- User-Agent Client Hints
User-Agent Client Hints are HTTP headers (the Sec-CH-UA family) that let a site request specific browser, platform, and version detail rather than reading it all from one passive string. They underpin UA reduction: the raw user agent is shrinking, and finer detail moves to opt-in hints. This page explains the model.
- Privacy-first analytics
Read device context without fingerprinting.
Sources and verification notes
- MDN — User-Agent Client Hints APIDefines how model hints are exposed and gated.
- Google Analytics Help — [GA4] Device dimensionsLists device model among the device dimensions.
Last reviewed 2026-06-24. Facts are checked against primary/official sources where available; uncertain specifics are marked “Data not yet verified” rather than guessed.