Week 1 2026: What the Numbers See in the Upcoming Matchups

Here's a finding we've published before and will keep leaning on: the opening weeks are our model's best stretch of the season — 56.6% against the spread across Weeks 1–3 (158–121 lifetime), its strongest window by a wide margin. Week 1 on its own sits at 56.8% (54–41). The reason is an information gap. September expectations are anchored to last season's narratives, while efficiency metrics carry over the offseason far better than reputations do. Week 1 is where the gap is widest.

So this is our data-first walk through the 2026 opening slate: where the model has real conviction, where it honestly shrugs, and why it lands where it does. This week, it shrugs a lot.

One note before the numbers. SpreadTrends is against-the-spread by default, and this article mixes both objectives, so every record below is labelled: ATS means the cover record, straight up means the win-loss record. They diverge more than you'd expect — Dallas closed last season 0-5 ATS but 1-4 straight up, and Minnesota's 4-0-1 ATS run included a push, not a tie.

Banner night in Seattle is a Super Bowl rematch

The season opens Wednesday night with the Patriots at the defending-champion Seahawks (Seattle -3, total 44.5) — the same two teams that played Super Bowl LX seven months ago, where Seattle won 29-13 as a 4.5-point favorite.

The model projects Seattle both to win (3★) and to cover (3★, its only spread projection above two stars all week), but the matchup card explains why neither number is higher: New England owns the league's best net pass EPA (+0.272 per dropback, rank 1), narrowly ahead of Seattle's +0.253 (rank 3). On net pass efficiency the order flips — Seattle +2.15 (rank 1) to New England +1.76 (rank 2). These are the two best passing-efficiency profiles in the NFL sharing one field on opening night, which makes a banner-night blowout the narrative read, not the numbers read.

Seattle closed 2025 at 5-0 both ATS and straight up through its title run; New England closed 4-1 straight up, 3-2 ATS. Their two meetings in the last three seasons produced one Seattle cover and one push — a two-game sample, which is all a head-to-head bar can honestly be worth here.

The Melbourne experiment

Thursday's 49ers–Rams game at the Melbourne Cricket Ground (Rams -3.5, total 48.5) is the NFL's first regular-season game in Australia, and it produces the week's cleanest split verdict: the model projects the Rams to win (4★, 76.9%) but the 49ers to stay inside the number (2★, 58.8%). A close Rams win, in other words.

The efficiency gap genuinely favors Los Angeles — net pass efficiency +1.36 to +0.44, net pass EPA +0.241 to +0.030, net rush efficiency +0.35 to San Francisco's -0.60 (rank 30). Recent form cuts the other way depending on which record you read: the Rams closed 1-4 ATS but 3-2 straight up, the 49ers 3-2 on both. Over three seasons the Rams have won four of six meetings.

The honest caveat writes itself: nobody — including our model — has historical data on what a 17-hour time-zone shift does to two West Coast teams. Every number here was learned on games played in the other hemisphere.

The conviction is in the winners, not the spreads

This is the shape of Week 1 2026, and it's unusual enough to be the story: the model has almost no spread conviction and a great deal of straight-up conviction.

Against the spread, nine of sixteen projections sit in the bottom tier (1★, under 56.3% projected), six are 2★, and exactly one — Seattle — reaches 3★. Nothing is rated 4★ or 5★. On the winner objective, the same slate produces three 5★ projections and five 4★ ones.

That is not a contradiction. The two objectives are separately trained models answering different questions, and a spread is designed to be a coin flip in a way an outright winner is not. When the market has priced a game well, "who wins" can be knowable while "by how much, relative to the number" is not. This week the market appears to have priced the spreads well and the model says so.

The three straight-up projections it will stand behind:

• Jets to win at Tennessee (5★, 82.0%). The week's single most confident projection, and it's on the road team in a game the market has at Titans -1.5. It is also the ugliest matchup on the board: these are the two worst net passing-efficiency profiles in the league, New York at -2.49 (rank 31) and Tennessee at -2.54 (rank 32), and they rank 32nd and 31st in net pass EPA. Tennessee is coming off a 23.5% straight-up three-year stretch, worst in the league. The model isn't saying the Jets are good; it's saying the Titans are worse.
• Chargers to beat Arizona (5★, 87.0%). The highest projected win probability of the week, against a Cardinals side that closed 2025 at 0-5 both ATS and straight up and has won 29.4% of its games over three seasons. Note the spread projection on the same game is only 2★ — the model likes Los Angeles to win far more than it likes Los Angeles to cover -9.5.
• Cowboys to beat the Giants on Sunday night (5★, 80.8%). New York's net rush efficiency is -1.02 (rank 31) and its three-year straight-up rate is 25.5%, the worst in the NFL. Dallas has won five of the last six meetings straight up. Worth flagging plainly, because it cuts against the model: Dallas closed last season 0-5 against the spread. The model is reading the Giants' profile, not Dallas's cover record.

Two places the model contradicts itself — and why that's the useful signal

Two games produce projections that cannot both be right:

• Jets at Titans. Spread projection: Titans (-1.5) to cover. Winner projection: Jets to win. Covering -1.5 requires winning by two.
• Bills at Texans. Spread projection: Bills (-1.5) to cover. Winner projection: Texans to win. Same impossibility.

This is what independently trained objectives look like when a spread is smaller than a single point of margin — "cover" and "win outright" collapse into nearly the same event, and any disagreement between the models becomes visible. In both cases the spread projection is 1★ (53.8% and 52.0%, barely off a coin flip) while the winner projection is 5★ and 3★ respectively. Read that as the spread side having nothing to say, not as two rival opinions of equal weight.

The Bills–Texans matchup is genuinely close on the numbers: Buffalo is second in net pass EPA (+0.265) and third in net pass efficiency (+1.41); Houston is fifth and eighth (+0.197, +0.78). Houston has won both meetings in the last three seasons.

Where recent form and efficiency argue

Three games this week set a team's recent cover record directly against its efficiency profile. The model sides with efficiency in two of them — and, usefully, not in the third:

• Packers at Vikings (Green Bay +1.5, ATS projection Packers 1★; winner projection Packers 2★). Green Bay lost its last five straight up (0-4-1 ATS); Minnesota won its last five (4-0-1 ATS). The model still leans Green Bay on the road, on a +1.21 net pass efficiency edge (rank 5) against Minnesota's -0.22. It leans quietly — neither projection clears two stars.
• Saints at Lions (Detroit -7, ATS projection Saints 1★; winner projection Lions 2★). New Orleans arrives on a 5-0 ATS close to 2025 (4-1 straight up); Detroit closed 2-3 on both. Detroit's efficiency is far better (+0.81 net pass efficiency to +0.15, +0.154 net pass EPA to -0.001) and its three-year straight-up rate is 69.1%, second-best in the league, against New Orleans' 39.2%. The model takes Detroit to win but wants the seven points.
• Broncos at Chiefs (Kansas City -2.5, ATS and winner projections both Chiefs 2★). Denver's efficiency profile reads clearly better — +1.03 net pass efficiency (rank 6) to Kansas City's -0.35 (rank 21), +0.179 net pass EPA to +0.042 — and Denver has won four of the last six meetings while Kansas City closed 2025 at 0-5 straight up. The model takes the Chiefs anyway, at modest confidence. Not every disagreement resolves toward efficiency.

The underdog question

There isn't one this week. The model's sixteen spread projections split exactly 8-8 between favorites and underdogs — 49ers, Saints, Buccaneers, Colts, Falcons, Panthers, Packers and Dolphins on the underdog side. In a week where nine of those projections are lowest-tier, an even split is roughly what a model with no view would produce, and we'd rather say that than dress it up.

The Ravens–Colts line is the sharpest example: Indianapolis +3.5 is projected to cover at 50.2%, the closest any projection on the board comes to a literal coin flip. The two teams rank first and second in the league in net rush efficiency (Baltimore +1.07, Indianapolis +0.67) and first and second in net rush EPA. The model can see they're similar and has nothing else to add.

Totals: where we hold our projections loosest

Our own research says totals are the hardest target we point the machine at, so read this section with that published humility. Nothing on the board reaches 5★, but four games reach 4★ — and, unlike the spreads, the totals model is willing to commit:

The most confident totals projection of the week is the Under in Melbourne — on the one game nobody has a precedent for. Note what that means alongside the spread board: the model reaches 4★ on four totals and never once on a spread. Whatever it thinks it can see this week, it isn't in the margins.

Two contextual notes, and a caveat about both. The week's highest total is Buccaneers–Bengals at 50.5, hung on two teams whose efficiency profiles are negative on both sides of the ball (Cincinnati's net pass efficiency is -1.06, rank 29) — the model leans Under, but only at 2★. The lowest is Jets–Titans at 39.5, set for those two grim passing profiles; the model leans Over there at 3★, which is worth flagging, because our research found the "low totals bounce over" pattern is the one that reverses out of sample. Nothing in that research supports the Over here; the model got there on its own.

The caveat is that neither number is actually extreme. This slate averages 45.4, so 50.5 sits about five points high and 39.5 about six points low — both inside the ordinary middle band, not the 7-plus tail where the high-total-under effect lives. Overall the slate splits 8 Overs, 8 Unders.

The six games this piece walks through

These are the matchups the analysis above is built on — the opener, the one in Melbourne, the two where our own objectives contradict each other, and two of the three 5★ winner projections:

All three objectives use the same 1–5 star scale, mapped from fixed confidence thresholds — 1★ is under 56.3% projected, 2★ 56.3–62.8%, 3★ 62.8–69.9%, 4★ 69.9–78.3%, 5★ 78.3% or better. A star therefore means the same thing on every card and in every week, rather than "the top of this week's distribution." Lines and totals are the numbers the model computed from on the evening of 2026-09-08 and will move before kickoff.

The remaining ten matchups — and every objective for all sixteen, with the full efficiency breakdown behind each one — are on the Predictions page, which is where the slate is published and graded each week.

What we won't pretend to know

More than half this slate sits in the model's lowest spread tier, and we've written it down rather than rounding it up into confidence. That's the honest shape of an NFL opening week — the league is engineered for coin flips, and Week 1 adds new coordinators, new quarterbacks, one Super Bowl rematch and one cricket ground to the pile.

Where we have conviction, the reasons are in the efficiency numbers above. Where we don't, we've said so, including the two games where our own objectives contradict each other. Every one of these projections will be graded in public either way, hits and misses alike — that's the deal.

Live matchup cards with the full efficiency breakdowns are on the Matchups page; the model's graded record, updated weekly all season, is on the Predictions page. Why September is the right time to trust the numbers: our accuracy-by-week analysis.