NFL Predictions 2026
Model as of · prices live
Our model's win probability on every game, side by side with the live Kalshi market price. The prediction is the call — the edge is where we disagree with the market.
59.6% accuracy · 3,266 games validated · Walk-forward, no hindsight · Methodology →
This Week's Predictions vs. Market
Every game on the slate — our win probability against the Kalshi price.
Model line = PMP Gridiron Edge favorite & spread. Market = Kalshi-implied probability for that side. Edge = model − market, in percentage points (a positive number means the model likes the favorite more than the market does). The market column fills in as Kalshi posts each game — Week 1 markets are live now.
Drafting this month? Same model, fantasy side.
The 2026 projection engine behind these numbers also runs a free fantasy draft desk inside Claude, ChatGPT or Cursor — the ranked board, the best available at your pick, sleepers and busts vs consensus ADP, and a who-do-I-draft call mid-draft. 8 tools, no account.
Open the fantasy draft assistant →Where the model disagrees
Our win probability minus the live prediction-market price, in percentage points. Biggest gaps first.
MIA @ LV
Model backs MIA · 62% vs 38% market
+25pp
MODERATE
BAL @ IND
Model backs IND · 58% vs 38% market
+21pp
MODERATE
BUF @ HOU
Model backs BUF · 57% vs 50% market
+7pp
MODERATE
NO @ DET
Model backs NO · 33% vs 26% market
+7pp
MODERATE
WAS @ PHI
Model backs WAS · 38% vs 33% market
+5pp
MODERATE
Informational only. Contracts can settle at $0 — verify the market and trade responsibly.
Season-Long Predictions
Live model calls against the market — top three from each board.
Model-ranked, all 32
PWR — points per game above an average team on a neutral field.
Patriots
+5.5
Contender
Seahawks
+5.5
Contender
Rams
+5.1
Contender
Biggest futures disagreements
Our simulated season win count against the listed contract.
Dolphins
Model 6.7 wins vs 4.0 line · Over
+39pp
STRONG
Browns
Model 7.8 wins vs 6.0 line · Over
+33pp
HIGH
Cardinals
Model 6.6 wins vs 4.0 line · Over
+32pp
STRONG
Playoff, conference & title gaps
Where our postseason probabilities part company with the market.
Browns
Playoff · model 36% vs 11% market
+25pp
STRONG
Dolphins
Playoff · model 23% vs 5% market
+18pp
STRONG
Titans
Playoff · model 24% vs 8% market
+16pp
STRONG
Mispriced MVP candidates
Model probability against the live contract price.
The three tightest races
Projected leader and how little separates them from second.
Falcons
NFC South · 0.2 clear of Buccaneers
-0.4
Tightest
Seahawks
NFC West · 0.4 clear of Rams
+5.5
Tightest
Packers
NFC North · 0.5 clear of Lions
+3.0
Tightest
This week, ranked by gap
Model win probability minus the market price, biggest first.
MIA @ LV
Model backs MIA
+25pp
MODERATE
BAL @ IND
Model backs IND
+21pp
MODERATE
BUF @ HOU
Model backs BUF
+7pp
MODERATE
Predictions by Week
Game-by-game model predictions for all 18 weeks of the 2026 regular season.
How our NFL predictions work
Every prediction starts from DAEPA — Defense-Adjusted EPA — which measures how efficiently a team moves the ball once you adjust for the strength of the defenses it faced. That efficiency rolls into a power rating, and the gap between two teams' ratings becomes a win probability through a Normal model of game margin.
We publish that number next to the live Kalshi contract price. When our probability and the market price diverge by 5 points or more — and the model agrees with the market's own line direction — we flag it as an edge. The backtest is walk-forward: 59.6% straight-up accuracy across 3,266 games, every call made with only the information available before kickoff.
We tried to fold a Madden-ratings prior into the opening weeks — we backtested it across 20 years and it carries a real, if modest, standalone signal. Fitted against this model as the baseline, though, the best weight moved Brier by 0.0007 and made results against the closing spread worse, so the fitted weight is zero and these predictions use no Madden input. Full method on the methodology page.
NFL Predictions — FAQ
How accurate are your NFL predictions?
In a walk-forward backtest of 3,266 NFL games from 2013–2024, our DAEPA model predicted winners straight-up 59.6% of the time with a 0.236 Brier score and a 0.664 log loss, with no market line among the inputs. Walk-forward means every prediction was made using only data available before that game, with no hindsight. Accuracy rises with the model’s own stated confidence, reaching 71.5% on the games where it flags an edge above 20%.
How are these NFL predictions made?
Each prediction starts from DAEPA (Defense-Adjusted EPA), which measures how efficiently a team moves the ball once you adjust for the quality of opponents it faced. That rolls into a power rating, which becomes a win probability via a Normal model of margin. We show that number next to the live Kalshi contract price — the prediction is the model call, and the edge is where the two disagree.
What is the difference between this page and the NFL schedule page?
The schedule page is the full weekly grid — every game, kickoff times, and links. This predictions hub leads with the model: this week's win probabilities vs. the market, the biggest edges, and the season-long futures. Use the schedule to browse; use this page for the call.
Do Madden ratings factor into the predictions?
No — and we measured that rather than assuming it. We backtested 20 years of Madden ratings (4,237 games) and found a real but modest standalone signal, strongest early. So we tried to blend it into these predictions for Weeks 1-4, fitting the weight across 947 games from 2010-2024 with our own preseason model as the baseline. The best weight improved Brier score by 0.0007 — noise — and made results against the closing spread worse. The fitted weight came out zero, so nothing here uses Madden. The signal is real on its own; it just isn't additive to a model already built on ~1.3 million plays.
Are the NFL predictions free?
The full-slate model lines, this-week predictions, power rankings, MVP market reads and the tier-graded game edges (moneyline, spread, total) are all free — no signup. Championship probabilities, player projections and win-total futures are fully open through Week 1; from September 15, 2026 those three show their top rows free and the full board is Pro at $14.99/month or $150/year.
Game edges are free. Always.
Model win probability against the Kalshi price on every game, every week — no signup. Futures, championship and projections are fully open through Week 1 too. Pro is the layer around them: edge alerts in Discord the moment a gap opens, the cross-market arb scanner, Thee Oracle, the commodity and crypto engines, and agent access to the same model over MCP.