Our own quarterback rating, built from play-by-play, adjusted for the defences faced and tested on six seasons it never saw. Recomputed after every completed week, with the full method below.
By C. B. Zakarian · Updated November 16, 2026
This page ranks every quarterback who has dropped back in the 2026 regular season, 61 of them so far, by the NFLAnalytic QB Power Rating: one number for how many expected points a quarterback adds per game compared with an average passer, built from what he does on every dropback and designed run, with the defences he faced taken out and small samples pulled toward what is known about him. It is a different tool from our QB rankings by EPA and CPOE, which show the raw season numbers; this rating asks what those numbers say about the rest of the season, and on 2020-2025, seasons it never saw while being built, it predicted that better than passer rating and better than raw EPA per dropback. The two part ways most this season on Geno Smith, 16 places apart; the reason, from the components, is at the end of the page.
Rating: expected points added per game above an average quarterback, at the league’s 39.3 dropbacks per team game. The three parts add up to it (rounding aside): passing (EPA per dropback, sacks and scrambles included), accuracy (completion percentage over expected) and rushing (designed runs). This table holds the 34 quarterbacks with at least 14 dropbacks per game their team has played, about a third of a team’s dropbacks, the workload of a starter for most of the season; it is the NFL’s 14-per-game rule for passing leaders, counted on dropbacks. Move compares each rank with the same rating computed through week 8, recomputed the same way every build. A tier break is drawn where two neighbours sit at least 0.9 points apart, half the typical uncertainty of one qualifier’s rating (1.8 points, one standard error); closer than that, the order is a lean. EPA & CPOE rank: the passer’s place on our EPA & CPOE table for the same weeks.
| Rank | Move | Quarterback | Team | G | Dropbacks | Rating | Passing | Accuracy | Rushing | Raw EPA/db / CPOE | EPA & CPOE rank |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | = | Brock Purdy | SF | 8 | 275 | +7.2 | +6.0 | +0.9 | +0.2 | +0.329 / +4.9 | 2 |
| 2 | = | Josh Allen | BUF | 8 | 300 | +6.9 | +5.5 | +0.6 | +0.9 | +0.289 / +4.0 | 5 |
| 3 | ▲ 1 | Joe Burrow | CIN | 8 | 337 | +6.8 | +5.0 | +1.4 | +0.4 | +0.278 / +9.7 | 1 |
| 4 | ▼ 1 | Drake Maye | NE | 9 | 337 | +6.2 | +4.6 | +1.4 | +0.2 | +0.219 / +6.7 | 3 |
| Tier break: 1.2 points between No. 4 and No. 5 | |||||||||||
| 5 | ▲ 4 | Jordan Love | GB | 9 | 385 | +5.0 | +4.7 | +0.4 | −0.1 | +0.213 / +1.1 | 11 |
| 6 | ▲ 1 | Dak Prescott | DAL | 9 | 400 | +4.4 | +3.9 | +0.2 | +0.3 | +0.215 / +0.1 | 13 |
| 7 | ▼ 1 | Trevor Lawrence | JAX | 8 | 260 | +4.3 | +3.3 | +0.1 | +0.9 | +0.258 / +2.8 | 6 |
| 8 | = | Jared Goff | DET | 8 | 309 | +3.9 | +4.1 | +0.3 | −0.5 | +0.244 / +1.3 | 7 |
| 9 | ▲ 1 | Patrick Mahomes | KC | 8 | 305 | +3.5 | +3.0 | +0.3 | +0.2 | +0.149 / +2.2 | 12 |
| 10 | ▲ 3 | C.J. Stroud | HOU | 8 | 332 | +3.4 | +2.9 | +0.1 | +0.5 | +0.198 / +1.8 | 9 |
| 11 | = | Caleb Williams | CHI | 6 | 238 | +2.8 | +3.5 | +0.3 | −1.0 | +0.226 / +6.4 | 4 |
| 12 | = | Kirk Cousins | LV | 9 | 358 | +2.5 | +2.3 | −0.1 | +0.2 | +0.202 / +0.2 | 14 |
| 13 | ▲ 1 | Matthew Stafford | LA | 9 | 383 | +1.9 | +2.2 | +0.0 | −0.3 | +0.107 / +0.0 | 16 |
| Tier break: 0.9 points between No. 13 and No. 14 | |||||||||||
| 14 | ▲ 1 | Sam Darnold | SEA | 7 | 248 | +1.0 | +0.8 | +0.1 | +0.1 | −0.005 / −2.5 | 25 |
| 15 | ▲ 1 | Jalen Hurts | PHI | 5 | 188 | +0.9 | +0.6 | +0.1 | +0.1 | +0.002 / −2.2 | 23 |
| 16 | ▲ 2 | Bo Nix | DEN | 9 | 341 | +0.7 | +1.1 | −0.3 | −0.1 | +0.085 / −1.1 | 19 |
| 17 | = | Jayden Daniels | WAS | 6 | 231 | +0.6 | +1.1 | +0.3 | −0.8 | +0.106 / +4.9 | 8 |
| 18 | ▲ 6 | Bryce Young | CAR | 8 | 358 | +0.5 | +0.4 | +0.1 | −0.0 | +0.082 / +1.3 | 15 |
| 19 | ▲ 2 | Kyler Murray | MIN | 6 | 199 | +0.1 | −0.2 | −0.0 | +0.3 | −0.023 / +2.0 | 17 |
| 20 | = | Michael Penix Jr. | ATL | 7 | 235 | +0.1 | +0.4 | −0.9 | +0.6 | +0.068 / −4.0 | 24 |
| 21 | ▲ 2 | Jacoby Brissett | ARI | 8 | 352 | −0.2 | −1.2 | −0.0 | +1.0 | −0.020 / −0.3 | 21 |
| 22 | = | Tyler Huntley | BAL | 5 | 135 | −0.3 | −0.8 | +0.3 | +0.1 | +0.015 / +0.9 | - |
| 23 | ▼ 4 | Geno Smith | NYJ | 9 | 331 | −0.4 | −0.3 | +0.5 | −0.6 | +0.102 / +4.2 | 10 |
| 24 | ▲ 4 | Jameis Winston | NYG | 6 | 199 | −0.5 | −0.6 | −0.9 | +1.0 | +0.003 / −6.1 | 28 |
| 25 | ▲ 2 | Tyler Shough | NO | 8 | 362 | −1.0 | −0.8 | −0.2 | −0.0 | +0.055 / −0.2 | 18 |
| 26 | ▼ 1 | Malik Willis | MIA | 8 | 278 | −1.3 | −0.6 | −0.4 | −0.3 | −0.020 / −4.8 | 27 |
| 27 | ▲ 2 | Aaron Rodgers | PIT | 8 | 334 | −1.4 | −1.2 | −0.4 | +0.1 | −0.051 / −1.9 | 26 |
| 28 | ▼ 2 | Baker Mayfield | TB | 6 | 184 | −1.8 | −1.6 | −0.2 | +0.0 | −0.220 / −1.5 | 29 |
| 29 | ▲ 1 | Daniel Jones | IND | 9 | 349 | −2.3 | −2.9 | −0.5 | +1.1 | −0.159 / −4.3 | 30 |
| 30 | ▲ 1 | Deshaun Watson | CLE | 8 | 328 | −2.5 | −1.7 | −0.2 | −0.5 | +0.043 / −0.1 | 20 |
| 31 | ▲ 1 | Cam Ward | TEN | 8 | 298 | −2.5 | −2.8 | −0.4 | +0.7 | −0.018 / −0.9 | 22 |
| 32 | ▲ 1 | Justin Herbert | LAC | 8 | 309 | −2.8 | −2.6 | −0.3 | +0.1 | −0.171 / −4.9 | 31 |
| 33 | ▲ 1 | Tanner McKee | PHI | 3 | 135 | −3.1 | −3.3 | −0.6 | +0.7 | −0.100 / −2.4 | - |
| 34 | ▲ 1 | Jalon Daniels | TB | 4 | 130 | −3.6 | −3.0 | −1.0 | +0.5 | −0.069 / −6.6 | - |
Tier breaks this week: after No. 4, after No. 13. Team: the club he dropped back for most often.
Biggest climb since the same rating through week 8: Bryce Young (No. 24 to No. 18); biggest drop: Geno Smith (No. 19 to No. 23).
Everyone else with at least one 2026 dropback, in rating order; the first column is his place among all 61. The rating needs no minimum sample, because a small one is pulled hard toward his prior, so a backup with a handful of dropbacks sits near where his past (or the average first-time quarterback) puts him. The last column says which prior he starts from.
| All-QB rank | Quarterback | Team | G | Dropbacks | Rating | Passing | Accuracy | Rushing | Prior |
|---|---|---|---|---|---|---|---|---|---|
| 5 | Lamar Jackson | BAL | 5 | 123 | +5.5 | +4.0 | +0.7 | +0.8 | last season carried |
| 6 | Jaxson Dart | NYG | 2 | 38 | +5.4 | +1.9 | −0.1 | +3.6 | last season carried |
| 11 | Marcus Mariota | WAS | 3 | 63 | +3.5 | +1.6 | +0.5 | +1.4 | last season carried |
| 21 | Tua Tagovailoa | ATL | 1 | 3 | +0.5 | +0.4 | +0.4 | −0.3 | last season carried |
| 23 | Mason Rudolph | PIT | 1 | 2 | +0.2 | −0.3 | +0.0 | +0.4 | last season carried |
| 26 | Drew Lock | SEA | 3 | 86 | −0.1 | +0.5 | −0.5 | −0.2 | last season carried |
| 31 | Gardner Minshew | ARI | 2 | 57 | −0.9 | −0.6 | −0.0 | −0.2 | last season carried |
| 35 | Tyson Bagent | CHI | 3 | 75 | −1.4 | −1.5 | −0.2 | +0.3 | last season carried |
| 37 | Carson Wentz | MIN | 3 | 80 | −1.8 | −1.1 | −0.6 | −0.2 | last season carried |
| 41 | Anthony Richardson | IND | 1 | 5 | −2.6 | −2.6 | −1.2 | +1.3 | last season carried |
| 42 | Case Keenum | CHI | 2 | 60 | −2.6 | −2.5 | −0.6 | +0.5 | first-time prior |
| 43 | Andy Dalton | PHI | 1 | 1 | −2.7 | −2.5 | −0.1 | +0.0 | last season carried |
| 44 | Joe Milton III | PHI | 1 | 48 | −2.7 | −2.2 | −0.5 | −0.1 | last season carried |
| 45 | Davis Mills | HOU | 1 | 2 | −2.8 | −1.7 | −1.1 | −0.0 | last season carried |
| 47 | Kenny Pickett | CAR | 2 | 7 | −2.9 | −1.3 | −0.3 | −1.4 | last season carried |
| 49 | Tommy DeVito | NE | 2 | 11 | −3.5 | −2.5 | −0.3 | −0.7 | two seasons back carried |
| 51 | DJ Uiagalelei | LAC | 1 | 8 | −3.6 | −2.7 | −1.0 | +0.1 | first-time prior |
| 52 | Stetson Bennett | LA | 1 | 4 | −3.6 | −2.8 | −0.8 | +0.0 | first-time prior |
| 53 | Fernando Mendoza | LV | 1 | 9 | −3.9 | −3.1 | −0.7 | −0.0 | first-time prior |
| 54 | J.J. McCarthy | NYG | 4 | 83 | −4.0 | −2.0 | −0.6 | −1.4 | last season carried |
| 55 | Quinn Ewers | JAX | 1 | 3 | −4.1 | −2.5 | −0.6 | −0.9 | last season carried |
| 56 | Jack Strand | ATL | 1 | 20 | −5.0 | −4.2 | −1.0 | +0.2 | first-time prior |
| 57 | Athan Kaliakmanis | WAS | 1 | 39 | −5.3 | −3.8 | −1.0 | −0.5 | first-time prior |
| 58 | Shedeur Sanders | CLE | 3 | 49 | −5.6 | −4.2 | −1.0 | −0.4 | last season carried |
| 59 | Max Brosmer | MIN | 1 | 8 | −5.6 | −5.3 | −0.5 | +0.2 | last season carried |
| 60 | Joe Fagnano | BAL | 2 | 47 | −7.1 | −4.7 | −1.0 | −1.4 | first-time prior |
| 61 | Cooper Rush | ATL | 2 | 49 | −7.5 | −5.9 | −1.1 | −0.5 | last season carried |
Quarterback here = listed at the position by nflverse, or 100+ dropbacks in the season. 13 other players threw from a trick play and are not ranked.
The question the rating answers is narrow on purpose: given every snap so far, how many expected points per dropback should we expect from this quarterback from here on, against an average defence? Everything below is there because it made that forecast better on past seasons, and the pieces that did not are named too.
The formula. Rating per dropback = 0.7 × adjusted EPA per dropback + 0.006 × adjusted CPOE (in percentage points) + adjusted designed-run EPA per dropback, each term measured against the league average and each shrunk toward a prior. Multiplied by 39.3 dropbacks, it becomes points per game.
Expected points added scores each play by how much it changed the offense’s expected points. A dropback is every play the play-by-play marks as a pass: completions, incompletions, interceptions, sacks and scrambles, credited to the quarterback (the same plays as our EPA & CPOE table). Sacks and turnovers are already inside this number at their real cost, which is why passer rating, which skips sacks, can rank a quarterback so differently. We also tried sacks and turnovers as separate parts with their own shrinkage, because turnovers are far less stable than the rest (see the stability table): splitting them did not help on the training seasons (error 0.1264 against 0.1260), so they stay inside EPA.
Completion percentage over expected compares each completion with how often the same throw (depth, distance, field position) is caught. It earned its place on the training seasons: added to EPA, it lowered the error of the forecast from 0.1270 to 0.1260 (how it fared on the held-out seasons is in the test section below). The weight, 0.006 EPA per dropback for each point of CPOE, came from the grid; at +1.2 points a game for a quarterback 5 points above expected, it is a nudge, not a second rating.
Scrambles are already dropbacks. Designed runs (option keepers, sneaks, draws for the quarterback, not kneel-downs) are added as EPA per dropback, so a running quarterback gets credit passer rating never gives him. Their stability puzzled us: split by alternate games, rushing value per dropback barely repeats (r 0.06), while how often a quarterback runs repeats strongly. Shrinking rushing by that split-half figure made the forecasts worse, so the grid set rushing’s prior strength directly: 0.01 × the split-half figure, about 43 dropbacks. Adding rushing lowered the held-out error on total quarterback EPA from 0.1311 to 0.1300.
Each quarterback-game is one row. One weighted regression on every 2026 row at once (weights = dropbacks) solves for a value per quarterback, a value per defence and a home-field term together, so a quarterback who has faced hard pass defences is credited for it and the defences themselves are measured on every quarterback they faced. This season the easiest pass defence to play against so far is the Arizona Cardinals’ (+0.124 EPA per dropback above average allowed) and the hardest the Minnesota Vikings’ (−0.189); the home term is −0.010 per dropback. Defence values are pulled toward average with the weight of 150 dropbacks, about four games of passes, so one bad afternoon does not make a defence look generous. Taking the adjustment out made the held-out forecast slightly worse (error 0.1275), so it stays.
A season is a few hundred dropbacks, and a few hundred plays carry a lot of luck. So each part is pulled toward a prior, with a strength measured, not guessed: split every quarterback-season of 2006-2019 with 200+ dropbacks into alternate games, compare the halves, and the correlation says how much of the difference between quarterbacks is lasting. From it comes k, the sample at which a quarterback’s own numbers and the prior count equally:
| Part | Half-vs-half r | Avg half sample | k | Used |
|---|---|---|---|---|
| EPA per dropback (all) | 0.50 | 248 | 247 | yes: passing |
| EPA on clean dropbacks | 0.48 | 248 | 265 | tested, not used |
| EPA lost to sacks | 0.39 | 248 | 396 | tested, not used |
| EPA lost to turnovers | 0.19 | 248 | 1,046 | tested, not used |
| CPOE (per scored attempt) | 0.45 | 209 | 259 | yes: accuracy |
| Designed-run EPA per dropback | 0.06 | 248 | 4,264 | yes, at 0.01 × k |
515 quarterback-seasons; k = half sample × (1 − r) / r. Turnovers repeat least of all (r 0.19): it takes about 1,046 dropbacks, more than a full season, before a quarterback’s turnover rate says as much about him as the prior does.
The prior itself: a returning quarterback starts from 85% of his final estimate from last season (85% squared if he sat out a year); a quarterback with no dropbacks in the two previous seasons starts from the measured average of such quarterbacks, 0.108 EPA per dropback and 3.2 points of CPOE below the league, because first-time and long-absent passers have been below average. Without the carried prior the held-out error rises from 0.1271 to 0.1301, the largest single loss of any piece.
We replayed 2007-2019 (2006 only seeds the priors): after weeks 4, 6, 8, 10, 12 of each season, the rating of every quarterback was compared with his EPA per dropback over the rest of that season (at least 50 more dropbacks), each miss weighted by those future dropbacks. The grid crossed shrinkage strength (0.5 to 2 × the measured k), the carried share of last season (0 to 95%), the defence pull (none, 50, 150, 500 dropbacks), a recency half-life (none or 6 weeks), the CPOE weight (0 to 0.009), the weight on EPA (0.55 to 1), split or pooled sacks and turnovers, and the rushing weight. The first run picked values at the edge of its range for three settings, so we widened those ranges on the training seasons only and ran it again; the winner did not change. Chosen: shrinkage 1 × k (the measured figure), 85% carried, defence pull 150, no recency weighting (it hurt: recent games are not more telling than early ones), CPOE 0.006, EPA weight 0.7. The EPA weight below 1 means that even after shrinkage, a quarterback’s EPA so far overstates how far from average he will be.
The settings were fitted on 2007-2019 and then run once on 2020-2025: 1,186 quarterback forecasts (each quarterback after weeks 4, 6, 8, 10, 12, against his EPA per dropback for the rest of that season). The other methods get a fair chance: each is turned into an EPA forecast by a simple linear fit on the same training seasons.
| Method | Error 2020-25 | Correlation 2020-25 | Rank agreement 2020-25 | Error 2007-19 | Correlation 2007-19 |
|---|---|---|---|---|---|
| NFLAnalytic QB Power Rating (passing parts) | 0.1271 | 0.47 | 0.48 | 0.1260 | 0.53 |
| EPA per dropback, season to date | 0.1322 | 0.36 | 0.47 | 0.1391 | 0.35 |
| Passer rating, season to date | 0.1297 | 0.40 | 0.43 | 0.1365 | 0.40 |
| CPOE alone, season to date | 0.1377 | 0.24 | 0.28 | 0.1439 | 0.25 |
| The EPA & CPOE page’s composite (equal-weight standard scores) | 0.1357 | 0.31 | 0.41 | 0.1400 | 0.34 |
Error: root mean squared miss in EPA per dropback, weighted by future dropbacks (lower is better). Correlation: between forecast and outcome, same weights. Rank agreement: the average Spearman rank correlation among quarterbacks with 100+ dropbacks before and after the cut-off. Raw EPA per dropback used as its own forecast, without the fit, misses by 0.2279.
On the six held-out seasons the rating has the lowest error and the highest correlation of the five methods. It does not win everything: on rank agreement it ties raw EPA per dropback (0.48 against 0.47); in 2025 raw EPA per dropback had the lower error; and the CPOE term, chosen on 2007-2019, did not help on 2020-2025, where the same rating without it would have missed by 0.1256 instead of 0.1271. We kept the setting the training seasons chose rather than tune it on the test. Correlation with the rest-of-season outcome by checkpoint (2020-2025):
| Cut-off | QB Power Rating | EPA per dropback | Passer rating |
|---|---|---|---|
| After week 4 | 0.50 | 0.38 | 0.41 |
| After week 6 | 0.47 | 0.33 | 0.41 |
| After week 8 | 0.48 | 0.39 | 0.40 |
| After week 10 | 0.46 | 0.36 | 0.39 |
| After week 12 | 0.44 | 0.39 | 0.37 |
Play-by-play credits the quarterback with the whole play. His receivers, his pass protection, his play-caller’s scheme and the down-and-distance he is handed all live inside his EPA, so a quarterback on a well-built offense is rated up and one on a poor one is rated down. A check from our own data shows how much: across 2006-2025, quarterbacks with 200+ dropbacks in two straight seasons for the same team kept their season rating with a correlation of 0.47 (430 cases); the 73 who changed teams kept it at 0.01, essentially nothing. Part of that is who moves (veterans past their best, backups getting a chance), but it is a plain warning: a large share of what any quarterback number measures belongs to the team around him. The rating still carries a mover’s prior with him, because the tested setting did; read a newly arrived starter’s early rating with that in mind (what a mid-season quarterback change costs, one starter against three). It also cannot see injuries, the score (late-game plays against soft coverage count fully) or anything that has not happened yet.
Passer rating uses completions, yards, touchdowns and interceptions per attempt; it ignores sacks, scrambles, runs, field position and the opponent (passer rating against QBR, passer rating calculator). Among this week’s 34 qualifiers, the largest gaps between the two orders, with the reason from the components:
Sources: plays from the public nflverse play-by-play (EPA and CPOE from the nflfastR models), names and positions from the nflverse weekly player stats, results from the nflverse game log. Every number on this page is computed by the site’s build from those files, nothing entered by hand; the full table is in qb-power-rankings-2026.json. It rates play on the field so far; it is not a projection of contracts, awards or anything to come.