This winter’s free-agent pitchers, graded by their pitches
The finding Of the 100 free-agent pitchers who threw 300+ MLB pitches in 2026, 9 posted results at least 15 points worse than their Pitching+ said their pitches deserved, and 10 at least 15 points better. History favors the grades: from 2025 to 2026, pitchers whose results trailed that much changed by +11.6 points of results the next season (90% interval +7.3 to +16.3), and those ahead by −9.7 (−12.6 to −6.7).
So what
- Results trailed the grades: Ryan Thompson (Pitching+ 120, RV+ 89); Tyler Zuber (Pitching+ 114, RV+ 86); Tim Hill (Pitching+ 120, RV+ 93); Blake Treinen (Pitching+ 117, RV+ 93). By past seasons, pitchers like these pitched closer to their grades the next year than their results suggest.
- Results ran ahead of the grades: Jacob Waguespack (Pitching+ 83, RV+ 113); Jordan Romano (Pitching+ 80, RV+ 105); Jordan Montgomery (Pitching+ 89, RV+ 114); Gregory Soto (Pitching+ 83, RV+ 108). Expect less than last season’s numbers.
- The best pitches on the market by Stuff+: Tim Hill (129, interval 128 to 131); Adrian Morejon (124, interval 123 to 125); Ryan Thompson (123, interval 121 to 125). The top starter by Pitching+ is Tarik Skubal (Pitching+ 117, interval 114 to 120).
- What this leaves out: health and medicals, age, workload, contract terms, and makeup. These are grades of the pitches thrown in 2026, not a full evaluation of a pitcher.
Every free-agent pitcher
Results against the grades
Why the grades, not the results
A season of results is noisy. Whether a hard-hit ball finds a glove, whether a two-strike pitch on the black gets the call, and a handful of bad outings in 60 innings all move a reliever’s numbers more than his talent does. The pitch grades look at something steadier: what every pitch did (its velocity, movement, and release for Stuff+) and where it went, in the count it was thrown (Location+ and Pitching+).
Free-agent talk usually starts from last season’s results. The test here asks how much those results say compared with the grades. Take every MLB pitcher with 500+ pitches in both 2025 and 2026 (324 pitchers), graded by models that never saw either season. Pitchers whose 2025 results trailed their Pitching+ by 15+ points (24 of them) saw their results change by +11.6 points the next season; those whose results ran 15+ points ahead (24) changed by −9.7; everyone in between, −1.6. Put both numbers in one model of next season’s results and last season’s Pitching+ carries a weight of 0.28 per point, against 0.09 for last season’s results.
That doesn’t make the grades a forecast on their own. Pitching+ predicted next season’s run value at r = 0.32, better than results (0.19) but far from certain, and one season is a small test. The gap is a reason to look harder, not a verdict.
Method and caveats
Who’s included. MLB’s free-agent list from the MLB Stats API as of November 14, 2026, pitchers only, leaving out players whose club option was exercised or who were claimed off waivers (they aren’t free). Of those, the 100 with 300+ graded MLB pitches in 2026 (35 starters, 60 relievers, the rest swingmen). 4 had already signed on that date. Signing status isn’t updated after it.
The grades. Stuff+, Location+, Pitching+, and results (RV+, actual run value) on one frozen scale: 100 is the average qualified MLB pitcher of the season before, 10 points is one standard deviation. They’re the same numbers as the leaderboard and player pages. How they’re built: Stuff+, Location+ and Pitching+. Models: Stuff+ v2.0.0 (validated in the backtest harness); Location+ v2.0.0 and Pitching+ v2.0.0 (experimental: validated before the backtest harness existed; its harness port is pending).
Intervals. 90%, from 1,000 resamples of each pitcher’s games (pitches in the same game move together). They cover sampling noise, not model noise: retraining the models with a different random seed moves a pitcher’s Stuff+ by 0.4 points on average and Pitching+ by 0.4, which is small next to these intervals.
The history test. Out-of-sample pairs: models trained on 2024 only grade 2025, and the outcome is 2026 run value, MLB pitchers with 500+ pitches in both seasons. The change is in RV+ points; intervals come from 1,000 resamples of the pitchers in each group. Part of the movement is plain regression to the mean: unusually good or bad results drift back for anyone. What the grades add is the direction, which pitchers’ results were too good or too bad for their pitches. One pair of seasons is a small sample, and the groups are smaller still.
What it can’t tell you. Health, medicals, and injury history; age and the aging curve; workload and role next season; contract cost; makeup and coachability. A pitcher whose results trailed his grades may have pitched hurt. Relievers’ samples are small, which is why their intervals are wide.
Code: baseball/free_agents/ · Data: Baseball Savant (Statcast) and the MLB Stats API. Sparvi is not affiliated with MLB or any club.