How Accurate Is TravelTime in the USA?

The most accurate drive time APIs in the US, benchmarked against Google

Accuracy is foundational for any location-based application, whether that’s a location search, a routing engine, an analysis platform, or an LLM working with location data. It’s what makes any decision about locations a reliable one.

Every location API provider understands how much accuracy matters for their users and promises a degree of reliability. But it’s hard to find out which one actually works best for your use case — and that’s why we’re sharing our new Accuracy Comparisons interactive benchmark.

We did the API benchmarking because we know that accuracy is important (and so you don’t have to).

We’ve compared six travel time and routing APIs — TravelTime, TomTom, HERE, Mapbox, Valhalla, and OSRM — across 97,000 US driving journeys, measured against Google.

Spoiler alert: TravelTime is in the top tier on every cut of the US data and has the lowest overall error of any provider tested.

TravelTime has the lowest error of any provider in the US

Our headline metric is RMSE (root-mean-square error): the typical difference, in seconds, between an API’s predicted travel time and Google API.

Out of the six providers we compared, TravelTime had the lowest RMSE in the US, at 507 seconds. That puts us ahead of TomTom (589s, 16% higher) and HERE (637s, 26% higher), and miles clear of Mapbox (993s), Valhalla (1,211s) and OSRM (1,475s). The average journey in the dataset is 47 minutes.

Why lead with RMSE? Because it’s the metric that’s hardest to hide behind.

RMSE is measured in absolute seconds and squares every error before averaging, so it surfaces the big misses that a simple average would smooth over. A routing API can look great on averages while quietly blowing up on a chunk of routes — and RMSE catches exactly that.

Having the lowest RMSE means TravelTime’s calculations have the fewest and mildest outliers of any provider tested.

Drive time API accuracy by US state

Routing engines need to adapt to different geographies, road networks, traffic profiles, and even driving styles. A strong national average can hide serious weak spots, and the state-by-state breakdown is where simple models get caught out.

TravelTime’s mean accuracy score in the USA is 87.1%, with its strongest performances in New Hampshire (93.6%), Mississippi (93.1%) and Rhode Island (93.0%).

But the more telling number is the floor. TravelTime’s weakest state — Indiana, at 78.9% — is still higher than any other provider’s weakest state. TomTom drops to 67.6% in Arizona, HERE falls to 52.1% in Montana, Mapbox hits 36.1%, Valhalla 20.4%, and OSRM bottoms out at 6.6% in Colorado.

Several drive time APIs post excellent numbers in well-mapped, traffic-rich states and then collapse in mountainous or rural ones.

TravelTime doesn’t have a Montana problem, or a Colorado problem, or an Alaska problem. Whichever state your users are in, the accuracy holds up.

Accuracy by time of day (peak vs off-peak)

Every route in the dataset was queried at three departure times (05:00, 13:00, 17:00) because this is where traffic and peak-time modelling matters.

TravelTime’s mean accuracy score moves less than four percentage points across the whole day: 88.8% in the early morning, 87.1% at midday, 85.3% in the evening peak.

Compare that with Mapbox, which swings from 79.9% at midday down to 64.5% at 5am, with a model that struggles once free-flow assumptions stop applying.

The open-source engines, Valhalla and OSRM, sit well below the leading pack at every hour, since neither model live traffic conditions.

TravelTime, TomTom and HERE form a clear top tier throughout the day.

Over-prediction vs under-prediction bias

Accuracy scores and RMSE tell you how far off a prediction is, but not which way it leans. Over-predicting means quoting users a longer trip than Google; under-predicting means a shorter one. Either bias, if systematic, quietly distorts whatever you build on top.

TravelTime is one of the most balanced providers tested: a median signed error of just +3.3%, with 60% of routes over-predicting and 40% under.

Contrast that with Mapbox and OSRM, which over-predict on 94% and 96% of routes respectively, both consistently leaning towards calculating journeys significantly longer than the Google API predicts.

Methodology: How we benchmarked the travel time APIs

Every number comes from the same dataset and the same formulas.

We sampled over 45,000 randomised origin–destination pairs across all 50 states, spanning a representative mix of urban and rural routes from 10 minutes to 3 hours long. Each was queried at three departure times through six location APIs plus Google.

After filtering out bad map-snaps, restricted-road artefacts, and routes any provider (Google included) couldn’t return, 97,534 clean routes remained.

Google’s predicted travel time is the reference point throughout. Not because it’s perfect, but because it’s the closest thing the industry has to a shared ground truth.

The full methodology, formulas and every cut of the data are on the Accuracy Comparisons page.

Beyond accuracy: What else to weigh when choosing a routing API

Accuracy is only one axis, and on its own it rarely decides which API a team actually ships with. By the time you’re in production with a routing API, a few other things matter just as much:

Performance at scale

An accurate response that arrives slowly doesn’t help when you’re computing a matrix of thousands of origins against thousands of destinations in real time. TravelTime is engineered for exactly that workload. See the performance methodology here.

Pricing that doesn’t punish growth

Per-request pricing means your bill scales linearly with usage. While this can work for low API usage, it starts to punish growth as you scale.

TravelTime’s fixed-price, unlimited-call model keeps costs predictable no matter how much you query.

Read more about the challenges of per-call pricing and why PAYG pricing breaks under AI workloads.

Caching and data rights

Several of the big location platforms restrict how long you can store their results, or whether you can cache at all. This rules out many location analytics and batch use cases before accuracy even enters the conversation.

TravelTime grants caching rights, so you can store and reuse results within your own systems.

Support that answers

Routing integrations are rarely plug-and-play, so developer documentation and support teams can play a critical part in their success.

TravelTime offers direct access to the people who built and maintain the API, whenever you need it, at no extra cost. No un-resolved tickets or account management blockers.

Benchmark the TravelTime API on your own routes

We’ve called the APIs, paid the bills and run the numbers. But we invite you to run your own comparison where it matters to you.

Get a free TravelTime API key and benchmark it on your own routes across the US and beyond.