NFL data for betting models

Quick answer: A betting model needs history to fit on, projections to start from, matchup and market context, and fast injury news. GridIron Data serves all of it from one REST API. Weekly results for every week since 2020 are on Basic ($9 a month or $59 a year). Pro ($29 a month or $249 a year) adds per-stat projections, defense vs position, spreads and totals, prop probabilities with fair odds, target share and an injury changes feed. It is data, not picks. Using it in a product that earns money needs Pro + Commercial ($49 a month or $490 a year).

History
2020–2026
Betting data
Pro
Price
$29/mo
Requests
50,000/mo

What does a betting model need, and which plan has it?

DataEndpointCoveragePlan
Weekly fantasy points and season totals (PPR, half, standard)/stats/{player_id}/{season}2020–2026Basic
Per-game box-score line (pass_yds, rush_att, rec, targets and the rest)box_score on /stats2025–2026Basic
Snap share by week, with a trailing trend/snaps/{player_id}2020–2026Basic
Actual vs projected points, week by week/deviations/{player_id}2025–2026Basic
Weekly fantasy point projections, with GridIron Data's own projection on each row/projections/{week}Current week on Free, every week on BasicFree / Basic
Per-stat projections (pass_yds, rec, rush_yds and the rest)projected_stats on /projections/{week}Every weekPro
Defense vs position: what each defense allows to QBs, RBs, WRs and TEs/defenseESPN box scores since 2025Pro
Spread, total, moneylines and implied team pointsodds on /scheduleRefreshed daily; the last line before kickoff is keptPro
P(over), P(under) and fair no-vig odds for a player stat line/propsQB, RB, WR, TE; 11 statsPro
Target share and carry share/usage2025 onwardPro
Injury status changes, polled or pushed to a signed webhook/injuries/changes, /webhooksUp to 14 days backPro
Depth charts and starters/depth-chartsDaily nflverse snapshotPro
Official game-day inactives, for checking what a backtest assumed/inactivesPublished after the gamesPro

The free plan covers player lookups and this week's projections only, so a model needs at least Basic. Basic allows 5,000 requests a month at 30 a minute; Pro allows 50,000 a month at 100 a minute. Free and Basic keys that call a Pro endpoint get HTTP 403 with reason: "pro_required", and the same key works after an upgrade.

What does a prop query look like in Python?

Send the player, the stat and up to 10 lines. The response has the projected mean and spread, the distribution used, a probability for each line and the model's walk-forward backtest for that stat.

import os, requests

API = "https://api.gridirondata.com/api/v1"
H = {"x-api-key": os.environ["GRIDIRON_API_KEY"]}  # a Pro key

r = requests.get(f"{API}/props", headers=H, timeout=30, params={
    "player_id": "Justin Jefferson#WR", "stat": "rec_yds", "line": "64.5,74.5"})
r.raise_for_status()
p = r.json()
print("mean", p["projection"]["mean"], p["projection"]["distribution"])
for line in p["lines"]:
    print(line["line"], "p_over", line["p_over"], "fair_odds_over", line["fair_odds_over"])
bt = p["model"]["backtest"]
print("backtest brier", bt["brier"], "base rate", bt["brier_base_rate"], "n", bt["n"])

The values the API returned for week 11 on November 19, 2026:

mean 66.16 gamma
64.5 p_over 0.4359 fair_odds_over 129
74.5 p_over 0.3441 fair_odds_over 191
backtest brier 0.2236 base rate 0.2465 n 10355

A Brier score below the base rate means the probabilities beat always guessing the historical over rate; the response also has a calibration table. The same response carries the week's game line (spread, total, moneylines) and the team's implied points, so one call gives you the market context too. Players who are OUT, on IR or on bye get p_over 0. The full walkthrough is in Price NFL player props with a probability API, and the endpoint reference is on the player props API page.

Can I backtest without look-ahead?

  • Closing lines. Each /schedule game keeps the last line captured before kickoff, so finished games show the line you would have bet into. odds is null when no line was captured.
  • Projections as they stood. Past weeks of /projections/{week} are not changed by today's injury report. GridIron Data's own projections freeze at kickoff, and as_of says when each run was built.
  • Who actually sat. /inactives lists the NFL's official inactives by week and team, published after the games, to check what your model assumed.
  • Stat corrections. Box scores are re-checked for 10 days after each game, so NFL stat corrections made in that window are reflected.

For a worked example of grading projections against actuals across the league, see Backtest fantasy projections against actuals. The history endpoints are mapped in Historical data for backtesting.

How fast do injury changes arrive?

GridIron checks ESPN's league-wide injury report every 2 minutes from 8:00 am to 1:00 am ET (every 10 minutes overnight), so a status change usually reaches /injuries/changes within a few minutes of ESPN posting it. Poll with ?since= and you get only what is new, with the old status, the new status, detected_at and the game it applies to. A Pro key can instead register one HTTPS webhook and have each batch pushed, signed with HMAC-SHA256; webhook calls don't count against your monthly requests. Details: Track NFL injury changes on game day.

What the data does not do

  • It gives data, not picks. Prop probabilities are a model estimate, not betting advice.
  • It does not sell or scrape sportsbook player-prop lines. Bring the line you want priced.
  • Game lines come from one provider (DraftKings, via the ESPN scoreboard), refreshed daily until kickoff. Each game has one line, the latest captured, not a history of line movement.
  • There are no live in-game stats or in-play odds.
  • Per-game box-score lines start in 2025. Seasons 2020–2024 have weekly fantasy points, season totals, snap share and matchup history, not per-game stat lines.
  • GridIron Data's own projections cover QB, RB, WR and TE; kickers and defenses aren't covered yet.

Some fields are licensed for internal benchmarking only. The commercial license says, word for word:

Field or endpointWhat you may do
Prop probabilities and fair odds from /propsInternal benchmarking only. Don't display or resell them.
Consensus projections: projected_points, projected_stats and the rest of /projections, /projections/ros, and the projected side of /deviationsInternal benchmarking only. No display, no republishing, and no training a commercial model on them.
Betting lines on /schedule (spread, total, moneylines, implied points)Use as a model input and for benchmarking. Don't republish them as an odds feed.

And the pricing page: "The consensus projections and prop probabilities are for internal benchmarking only under every plan; GridIron Data's own projections may be shown to users with Pro + Commercial."

Sports betting is legal only in some places and only for adults, 21 and over in most US states. Follow your local laws. If gambling stops being fun, call 1-800-GAMBLER.

Pro or Pro + Commercial?

Pro ($29 a month or $249 a year) covers personal, internal and free non-commercial use: your own model, your own bets, research you don't sell. Pro + Commercial ($49 a month or $490 a year) has exactly Pro's data access plus the license. You need it when the data feeds something that earns money, through subscriptions, ads, sponsorships, affiliate links or paid tiers. Under it you may "Publish your own derived outputs (projections, probabilities, rankings, analysis) to paying users" and train your own models on the historical data, except the benchmark-only fields above. It is for companies under $250,000 a year in revenue; larger companies need Enterprise. Read the Commercial License and compare every plan on the pricing page.

Frequently asked questions

Does GridIron Data give betting picks?

No. It sells data: stats, projections, game lines, defense vs position, injury changes and model probabilities for a line you send. Prop probabilities are a model estimate, not betting advice.

How far back does the NFL data go?

Weekly fantasy points, season totals, snap share and matchup history go back to 2020. Per-game box-score stat lines, defense vs position and target share start in 2025.

Which plan do I need for a betting model?

Basic ($9 a month or $59 a year) covers the 2020-2026 history and every week's projections. Pro ($29 a month or $249 a year) adds per-stat projections, defense vs position, game lines, prop probabilities, target share and injury changes. Pro + Commercial ($49 a month or $490 a year) adds the license to sell what you build.

Can I show GridIron's prop probabilities in my app?

No. The commercial license lists them as internal benchmarking only: "Don't display or resell them." Probabilities from your own model are yours to publish to paying users with Pro + Commercial.