We may earn a commission from links on this page.
As the average running watch gets increasingly high-tech, there’s a new normal for pre-race rituals: Staring at your watch the week before a big event, hoping the algorithm has good news for you.
Both Garmin and Strava offer race time predictions, but how accurate are they, really? I put each platform to the test during the Brooklyn Half Marathon on May 16, 2026. Here’s what I experienced, and what I think every runner needs to know about what these tools actually measure.
How Garmin’s Race Predictor works
Garmin’s Race Predictor has been a fixture on its mid-range and advanced running devices for more than a decade. The feature provides estimated finish times for the 5K, 10K, half marathon, and full marathon, and it works primarily by translating your estimated VO2 max into race pace equivalents. Garmin says it also uses personal data (age, gender) and recent training history to moderate short-term fluctuations.
Of course, this sort of model assumes you’ll execute a perfect race. That means optimal pacing, ideal weather, full taper, proper fueling, mental fortitude, and so on. And while Garmin does display a heat or altitude indicator on the VO2 max widget when conditions affect that estimate, that indicator does not carry over into the Race Predictor itself. In this way, it more accurately predicts your aerobic ceiling, and less so your true expected finish time.
On higher-end devices like the Forerunner 965 and 970, Garmin offers a more sophisticated “Course and Weather-Specific Race Predictor” when a race is entered into the Garmin Connect calendar. This can apply course elevation and environmental adjustments, like, say, race-day heat (consider this foreshadowing).
For context, I ran this past race with my Garmin Forerunner 970 as my primary watch. (I’m currently working on a comparison of race-day performance between the 970 and the Amazfit Cheetah 2 Pro).
How Strava’s performance predictions work
Strava’s Performance Predictions feature is newer, having launched in April 2025. Rather than routing everything through a theoretical VO2 max estimate, Strava’s system uses AI and “real activity data”—your own and that from other runners. This activity data allegedly includes over 100 data attributes, including a runner’s all-time activity history, recent training load, top performances, and the performances of other Strava users with similar training histories.
Because each race distance is calculated independently, Strava argues its system achieves greater precision at each distance, rather than extrapolating one metric across all of them. The model generates a new prediction after each run upload and requires at least 20 run activities within a rolling 24-week window.
In my circles, most runners find Strava’s predictions to be a bit volatile and “random” compared to Garmin’s. This makes sense, as someone who has watched my prediction go dramatically up or down after a single bad (or exceptional) run. On the flip side, Strava acknowledges that the model gives significant weight to all-time history, which can cause it to lag for runners returning from injury or coming off a long break. One notable limitation: Performance Predictions do not account for terrain or altitude. They assume a flat course, similar to a track.
What Garmin and Strava predicted before my race
Garmin predicted: 2:00:51. This would have been a personal record. In retrospect, it offers a useful window into how Garmin’s model behaves. The prediction almost certainly reflected strong recent VO2 max readings from training runs, translated into an idealized race-day outcome.
Strava predicted: 2:10:34. This is a much more conservative number, slower even than my last official half marathon from last September (2:05). Given that Strava leans heavily on historical performance data, including all-time best efforts, this prediction may have been anchored to all my easy training run paces, rather than race-effort data, or it may have reflected a training block that didn’t include much high-intensity running at half-marathon-specific effort.
The range between the two predictions—nearly ten full minutes!—is itself a story. For context, at a 10K earlier in May, Garmin predicted 54:04, while Strava came in at 58:14, a difference of over four minutes. (That race was ultimately run extra conservatively due to a knee injury, so I have no interesting results for you there.) But the pattern is telling: Garmin skews optimistic, and Strava skews conservative.
My results: Smack dab in the middle
Because this was a real-world test, I want to note the real-world conditions that affected my time. The Brooklyn Half had a gorgeous course advantage baked in, where the full second half is a net downhill. Many runners target personal bests at this race specifically because of it.
Unfortunately, racing in May weather is unpredictable, and race day was a scorcher compared to training. The temperature was at least ten degrees Fahrenheit warmer than any of my runs in the lead-up—a significant variable for a runner who is quick to fold in the heat. Plus, the downhill portion offered no cloud cover. I made some water station stops in a deliberate effort to manage my heart rate, even at the cost of pace. For any runner who has pushed too hard in the heat before, you know how the mental calculus shifts: finishing healthy outweighs finishing fast.
Credit: Meredith Dietz
My final time was 2:04:49. This number splits the difference between Garmin and Strava in a suspiciously neat way. Garmin’s prediction was 3 minutes and 58 seconds too fast. Strava’s prediction was 5 minutes and 45 seconds too slow. So, Garmin was the more accurate of the two, but neither prediction was wrong in a way that would cause a runner to make a catastrophically bad pacing decision.
Remember, Garmin’s Race Predictor is engineered to tell you what your aerobic system is theoretically capable of under perfect conditions. For short distances like the 5K and 10K, that ceiling and reality could be pretty close. For the half-marathon and marathon, the gap widens—and it widens dramatically when race-day conditions deviate from the calm, cool training runs that shaped your VO2 max estimate. Runners who use Garmin’s prediction as a pacing target without accounting for heat, course difficulty, or their own racing readiness risk going out too fast and paying for it in the second half.
Strava’s heavy weighting toward historical data and comparable athletes may cause it to underestimate a runner who is currently in strong shape but hasn’t recently logged race-effort results for Strava’s algorithm to learn from. If you train mostly at easy paces and rarely race, Strava may not have enough signal to recognize your current ceiling. Plus, Strava’s own community has noted that predictions can swing substantially based on a single run, which makes it harder to build race-day confidence around a moving target.
The bottom line
Garmin estimates your aerobic potential under ideal conditions; Strava estimates what a runner with your training history has realistically achieved. Both approaches have blind spots, and both will mislead you if you treat their output as gospel.
For me, Garmin came closer to the actual finish time, but at the same time, it was the more dangerous prediction to follow on a hot day if I hadn’t erred on the side of caution. Whatever your predictions say on race morning, remember to consider the forecast, know the course, and allow for a bit of a buffer.
