Best running watches for marathon training

Race day is the part that is already solved

Marathon watch roundups almost always open with battery life, and for a road marathon that is the wrong lead. Below is every watch on this page with the endurance figure its own maker publishes, the satellite mode that figure refers to, and how many times over it covers a five-hour marathon.

Two honest caveats about that table. These figures are not like-for-like: Garmin publishes a separate multi-band number for the 965 and the 265 and we quote it, while the headline figures we carry for the 970 and the 570 are GPS-mode numbers, and COROS publishes no dual-frequency figure for the PACE 4 at all. Comparing across rows compares different satellite modes, which is exactly the trick that makes spec sheets misleading without lying.

And second: none of it matters much for 26.2 miles. The tightest watch here still finishes two marathons on one charge. Battery is a real argument for an ultra, and we make it on the ultrarunning page. It is not an argument for a marathon.

What actually decides a marathon block

A sixteen-week build is roughly sixty sessions, and the ones that matter most are the threshold and marathon-pace runs. Every modern watch will hand you a threshold pace and a threshold heart rate, and modern training plans are written around exactly those numbers. So the question worth asking is not how long the battery lasts. It is whether the number is right.

There is exactly one published answer, and one of the watches on this page is in it.

The only watch here that has been measured

A 2025 study in Frontiers in Physiologyput 100 recreational runners through a treadmill graded exercise test with fingertip blood lactate sampling, then had them run each device's own on-device threshold protocol. The Garmin Forerunner 265 was one of three devices tested, on 23 participants.

• It produced an estimate 65.2% of the time. Fifteen successful estimates out of 23 attempts, with the protocol followed correctly. A third of the tests simply did not yield a number.
• Threshold heart rate was out by 11.44 bpm on average — a 7.15% mean absolute percentage error, correlating with the laboratory value at r = 0.67.
• Threshold pace was out by 2.17 km/h on average, a 25.78% error, and significantly overestimated (p < 0.01).

Overestimated is the word that should change your training. If you take the watch's threshold pace at face value and run your tempo sessions at it, the published evidence says you are more likely to be running them too fast than too slow — which is the single most common way a marathon build goes wrong.

The authors also ran equivalence testing, the stricter question of whether watch and laboratory can be treated as interchangeable. None of the three devices met it. We work through the full study, its limits and its small per-device samples in what a lactate threshold estimate is worth.

The uncomfortable consequence for this ranking

The Forerunner 265 finishes fifth on our rubric and it is the only watch on the page whose threshold estimate anybody has checked. The four watches above it have no threshold measurement at all, and neither does the PACE 4 — the only COROS ever tested was the previous PACE 3, which produced an estimate 47.1% of the time.

We are not going to resolve that by promoting the 265. The rubric weights evidence, hardware, metric coverage, endurance and cost per capability the same way on every page of this site, and being the device that was measured badly is not the same as being the best device. But it does mean this: nobody reading this page should treat a watch threshold number as a laboratory result, on any of these five.

Where the money actually goes

Cost per hour of GPS recording is arithmetic we can do from the live price and the maker's own stated hours, and it separates these watches more honestly than the feature lists do. The Forerunner 970 works out at $28.58, the Forerunner 965 at $31.58, and the COROS PACE 4 at $6.07 — remembering that the COROS figure is an All-Systems number and the Garmin ones are not.

The gap between the flagship and the mid-range is $303 today. For that you get maps, 32 GB of media storage instead of 8, an ECG app, and a speaker. None of those makes the training data better.

The long-run details nobody warns you about

Two practical things do change during a marathon build. The first is music: Garmin's own tables show that playing audio roughly halves recording time — the 965 states 19 hours in All + Multi-Band and 8.5 with music, and the 265 states 14 and 6. Both still finish a marathon comfortably, but a six-hour trail long run with headphones is a different calculation. The music ranking has the full storage and battery table.

The second is heart rate. Wrist optical sensors are the weakest link in a marathon block because the sessions that matter most are the ones with sharp intensity changes, and that is precisely where wrist sensors fail. If your tempo sessions show a heart rate that lags the effort by a minute, the fix is a chest strap rather than a new watch — that argument, with the study behind it, is in fix your heart rate data.

Running power is the third number people build marathon plans around, and it deserves the same skepticism as the threshold estimate: we set out what it actually measures, and what has and has not been validated, in what running power actually measures.

How to choose for a marathon block

• Ignore battery unless you are running longer than about eight hours. Every watch here clears a marathon twice over in its accurate mode.
• Decide whether you want maps. That, not accuracy, is the real flagship-versus-mid-range decision. We take it apart in do you need maps.
• Budget for a chest strap before you budget for a better watch. It costs a fraction as much and it is the only part of the chain with peer-reviewed evidence behind it.
• Treat the threshold number as a starting hypothesis. Check it against a race result or a hard effort you know, and be willing to move it down.

Every score on this page comes from one published rubric applied to every device on the site. The weights, and what we refuse to claim, are at how we rank.