NFL Predictions Week By Week: Where Models Fade The Market
By Jake Hari
August 12, 2026
NFL Predictions Week By Week: Where Models Fade The Market
Updated August 12, 2026 · 14 min read · by Jake Hari
This is the page we rebuild every week of the season. The idea is simple: take the live win probability that real money has assigned to every game on the slate, run the same eight AI models over the same fetched facts without letting them see a single price, and publish both numbers side by side. Where they agree, there is less to argue about. Where they split, you have found the part of the slate worth an extra ten minutes before lineups lock.
The Quick Answer
The panel's loudest disagreement with the market is in Las Vegas. Traders make the Raiders a 64.5-cent home favorite over Miami; the eight models average 49.8%, which is a coin flip, and not one seat priced Las Vegas above 55%. The second loudest is Monday night, where Kansas City is a 56.5-cent favorite and the panel puts the Chiefs at 45.4% with Patrick Mahomes carrying a live Questionable tag off ACL and LCL surgery. The one game where the panel sits well ABOVE the price is Indianapolis, at 36.5 cents against a 46.8% blend, the board's biggest gap in a home team's favor. The full 16-game board, every seat's number, the three games both sides call a coin flip, and the DFS read on each game environment are below.
New to event markets? Prediction Markets For DFS Players covers the mechanics in plain language; this page is the weekly game-by-game layer on top of it.
Prices on this page are re-read from the exchange every time we rebuild the board, and the stamp under each table tells you exactly when. Opening-week prices move on quarterback news, and four teams on this slate had an unresolved quarterback picture when these numbers were generated (Cleveland, Atlanta, Las Vegas and Arizona), with Kansas City waiting on a surgically repaired knee, so a number that looks stale probably is.
Free: The Weekly PM Market Brief — the panel's graded record, the week's biggest market-vs-model gaps, and what's spiking next. One email, Sundays, from the signup box on this page.
The Board
Every game of the 2026 opening week, home team listed second. One caveat on that: the Rams are the designated home team against San Francisco, but the game is in Melbourne, so nobody actually has a home field. "Market on home" is the live price of the home team's winner contract in cents, which reads directly as a percentage. "Panel blend" is the simple mean of eight independent model estimates, each made without seeing any price. The gap column is the panel minus the market, so a negative number means the models think the market is too high on the home team.
| Game (Home Team Second) | Kickoff | Market on home | Panel blend | Gap |
|---|---|---|---|---|
| New England At Seattle | Wed Sep 9, 8:20 PM | 65.5¢ | 56.8% | -8.7 |
| San Francisco At L.A. Rams | Thu Sep 10, 8:35 PM | 63.5¢ | 61.0% | -2.5 |
| Atlanta At Pittsburgh | Sun Sep 13, 1:00 PM | 59.0¢ | 59.5% | +0.5 |
| Baltimore At Indianapolis | Sun Sep 13, 1:00 PM | 36.5¢ | 46.8% | +10.3 |
| Buffalo At Houston | Sun Sep 13, 1:00 PM | 49.0¢ | 52.3% | +3.3 |
| Chicago At Carolina | Sun Sep 13, 1:00 PM | 41.5¢ | 43.4% | +1.9 |
| Cleveland At Jacksonville | Sun Sep 13, 1:00 PM | 76.0¢ | 78.5% | +2.5 |
| New Orleans At Detroit | Sun Sep 13, 1:00 PM | 74.5¢ | 69.4% | -5.1 |
| N.Y. Jets At Tennessee | Sun Sep 13, 1:00 PM | 57.0¢ | 53.5% | -3.5 |
| Tampa Bay At Cincinnati | Sun Sep 13, 1:00 PM | 64.5¢ | 57.0% | -7.5 |
| Arizona At L.A. Chargers | Sun Sep 13, 4:25 PM | 81.5¢ | 77.5% | -4.0 |
| Green Bay At Minnesota | Sun Sep 13, 4:25 PM | 52.0¢ | 52.1% | +0.1 |
| Miami At Las Vegas | Sun Sep 13, 4:25 PM | 64.5¢ | 49.8% | -14.7 |
| Washington At Philadelphia | Sun Sep 13, 4:25 PM | 66.0¢ | 70.3% | +4.3 |
| Dallas At N.Y. Giants | Sun Sep 13, 8:20 PM | 43.0¢ | 42.8% | -0.2 |
| Denver At Kansas City | Mon Sep 14, 8:15 PM | 56.5¢ | 45.4% | -11.1 |
Prices as of August 12, 2026, 7:11 PM ET. Model estimates generated the same day, price-blind.
Every Seat's Number
The blend hides the argument. Here is what each of the eight models put on every game after a revision round in which each seat read the other seven's reasoning without knowing which model wrote what.
| Game | ChatGPT | Fable | Opus | Sonnet | Gemini | GLM | Kimi | DeepSeek | Blend |
|---|---|---|---|---|---|---|---|---|---|
| New England At Seattle | 56% | 57% | 56% | 56% | 58% | 55% | 58% | 58% | 56.8% |
| San Francisco At L.A. Rams | 58% | 59% | 62% | 60% | 61% | 62% | 64% | 62% | 61.0% |
| Atlanta At Pittsburgh | 59% | 63% | 60% | 61% | 56% | 60% | 62% | 55% | 59.5% |
| Baltimore At Indianapolis | 47% | 47% | 45% | 48% | 49% | 45% | 45% | 48% | 46.8% |
| Buffalo At Houston | 51% | 52% | 53% | 51% | 54% | 52% | 52% | 53% | 52.3% |
| Chicago At Carolina | 43% | 42% | 44% | 43% | 42% | 42% | 46% | 45% | 43.4% |
| Cleveland At Jacksonville | 76% | 76% | 83% | 80% | 78% | 76% | 75% | 84% | 78.5% |
| New Orleans At Detroit | 66% | 70% | 76% | 72% | 67% | 65% | 71% | 68% | 69.4% |
| N.Y. Jets At Tennessee | 52% | 55% | 55% | 53% | 52% | 51% | 55% | 55% | 53.5% |
| Tampa Bay At Cincinnati | 56% | 58% | 57% | 55% | 60% | 55% | 58% | 57% | 57.0% |
| Arizona At L.A. Chargers | 74% | 79% | 79% | 77% | 81% | 76% | 76% | 78% | 77.5% |
| Green Bay At Minnesota | 52% | 51% | 52% | 52% | 51% | 52% | 54% | 53% | 52.1% |
| Miami At Las Vegas | 51% | 52% | 52% | 46% | 46% | 48% | 48% | 55% | 49.8% |
| Washington At Philadelphia | 68% | 68% | 71% | 70% | 74% | 68% | 72% | 71% | 70.3% |
| Dallas At N.Y. Giants | 42% | 42% | 44% | 45% | 44% | 41% | 42% | 42% | 42.8% |
| Denver At Kansas City | 44% | 48% | 47% | 44% | 43% | 42% | 47% | 48% | 45.4% |
Every seat on this panel is graded against real market settlements — records to date: Gemini 86% on 4,228 graded calls · GLM 81% on 2,318 graded calls · Kimi 83% on 2,296 graded calls · DeepSeek 80% on 2,343 graded calls. Recomputed daily; the full scoreboard is public.
These are model estimates, not predictions of fact and not financial advice. Every number here gets graded in public once the games settle, and you can read the running results on the full graded scoreboard.
More live boards from the same panel: NFL Week 1 market board (Chargers 81.5¢) · NFL rushing yards leader (Bijan Robinson 14.5¢) · NFL MVP 2026 (Josh Allen 10.5¢) · Prices fetched August 12, 2026.
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $178M traded
- 2027 Pro Football Champion · $60M traded
- 2028 U.S. Presidential Election winner? · $59M traded
- 2028 Republican presidential nominee · $56M traded
- Pro Baseball Champion · $52M traded
- More tech layoffs in 2026 than in 2025? · $31M traded
Every market above links to our full AI model verdict; browse them all on the Stokastic prediction markets hub, and see how every settled call actually scored on the full graded scoreboard.
Las Vegas Raiders: The Panel's Biggest Fade
The Raiders are the week's cleanest disagreement. The market has Las Vegas at 64.5 cents at home against Miami. The panel's average is 49.8%, and the highest number any of the eight seats gave the Raiders was 55%.
The case against the price is arithmetic. Las Vegas went 3-14 last season and scored 241 points, the fewest of any team playing in the opening week, while allowing 432. Miami was bad too, 7-10 with a minus-77 differential, but that is a hundred-plus points of separation in Miami's favor. The market appears to be paying for home field and for the assumption that Kirk Cousins steadies an offense that a rookie No. 1 pick, Fernando Mendoza, may eventually take over.
The models were not willing to pay much for either. Kimi moved toward the Raiders in the revision round and still landed at 48%, explaining the move this way:
Miami's room (Willis/Ewers/Miller) is no more settled than Vegas's, and if Cousins opens for the Raiders, home field plus QB stability roughly closes the -191 vs -77 differential gap. — Kimi
That is the bull case for Las Vegas at its strongest, and it produces a number below 50. Gemini went the other way entirely, cutting the Raiders from 61% to 46% after re-reading the fetched record, calling it a correction "on their abysmal 2025 differential and uncertain QB situation."
Patrick Mahomes And The Monday Night Question
Kansas City is the board's most interesting price because both sides of it are defensible. The Chiefs went 6-11 last season, which sounds like a collapse, but their point differential was plus-34. That is the profile of a team that lost a pile of close games, and the reason is on the injury report: Mahomes tore the ACL and LCL in his left knee in Week 15 and had surgery in December.
Reporting in late July had him near full clearance about seven months post-surgery, ahead of the typical nine-month timeline, and the plan was to be ready for the opener barring setbacks. He is also, as of today's league injury report, listed as Questionable. Denver arrives as last season's No. 1 seed in the AFC, 14-3 with a plus-90 point differential.
The panel split on exactly how much that Questionable tag is worth, and the split is the content. Opus cut Kansas City from 51% to 47%:
I had treated Mahomes as effectively cleared because of the ahead-of-schedule reporting; the card shows a live Questionable on today's report, and an ACL/LCL return is exactly where Questionable carries real snap-count risk rather than paperwork. — Opus
Sonnet moved the opposite direction, from 38% up to 44%, on the grounds that the reporting was stronger than the designation implied. Both seats are reasoning from the same two facts and landing three points apart, which is roughly the honest width of this game.
Indianapolis: The Home Underdog The Panel Prices Near Even
Both double-digit gaps above run against home favorites. Indianapolis runs the other way, and by the widest margin on the board in a home team's favor: the market prices the Colts at 36.5 cents against Baltimore, the panel's blend is 46.8%, and all eight seats sat between 45% and 49%.
The reason is that the Colts outscored their opponents by 54 points last season while Baltimore's differential was plus-26, and both teams finished 8-9. One plausible read is that the market is weighting Baltimore's quarterback edge more heavily than the two rosters. Opus, the lowest seat on Indianapolis in the first round at 42%, moved up after reading the other seven:
I leaned entirely on Lamar, but Indianapolis had the better 2025 differential (+54 to +26) and is home — I still like Baltimore, just less. — Opus
Baltimore's own opening week is complicated up front. Two defensive tackles, Travis Jones and John Jenkins, are listed out, along with linebacker Teddye Buchanan. The panel still makes Baltimore the marginal side at 53.2%, which is a much smaller edge than the one the market is charging for.
The Three Games Both Sides Call A Coin Flip
Three games came back from a fully enriched data card with the panel inside three points of even money, and on all three the market is close to the same place. That is a finding rather than a failure, because in each case the seats named the specific factors that cancel:
- Buffalo At Houston, Panel 52.3% Against 49.0 Cents. Two 12-5 teams whose 2025 differentials are seven points apart, plus-116 and plus-109. ChatGPT's first-round board filed it with the matchups where "small injury and venue edges mostly cancel"; Sonnet then raised Houston from 47% to 51% on the grounds that between teams this even, "home field should be the tiebreaker."
- Green Bay At Minnesota, Panel 52.1% Against 52.0 Cents. Minnesota named Kyler Murray its Week 1 starter on August 11 and carries no injury designations at all, which is worth something; Green Bay counters with continuity at quarterback and a slightly better 2025 differential, plus-31 to plus-11.
- N.Y. Jets At Tennessee, Panel 53.5% Against 57.0 Cents. Both teams finished 3-14, the Jets at minus-203 and the Titans at minus-194. When two rosters are this evenly matched at the bottom, home field is close to the only signal left.
Las Vegas belongs in that company on the panel's numbers, at 49.8%. It leads this article instead because the market is fifteen points away from agreeing. Priced boards get sharper than this in October; in August, a slate where six of the sixteen games sit within five points of a coin flip on the panel's numbers is what the data supports.
What This Board Means For Your DFS Lineups
Win probability is not a projection, but the shape of a game tells you what kind of lineup it feeds. Three reads worth carrying into your Week 1 builds:
The leaky-defense games are where the points are. Dallas at the Giants pairs a Cowboys team that scored 471 points and allowed 511 last season, the worst defensive number on this board, with a Giants team whose own differential was minus-58. Tampa Bay at Cincinnati puts Joe Burrow at home. New Orleans at Detroit is a mismatch on paper, and Detroit is missing three safeties, Brian Branch, Kerby Joseph and Christian Izien, all listed out.
The weak-offense side of the board is unusually deep this year. Cleveland has not named a quarterback and scored 279 points last season. Arizona is expected to open with a veteran whose 2025 starts went 1-11. Las Vegas scored 241. The Jets and Titans scored 584 points between them last season, both finishing 3-14. Those are the offenses to be underweight in tournament builds unless the price forces your hand.
The Melbourne game is a roster-attrition spot, not a home game. San Francisco plays the Rams at Melbourne Cricket Ground, a neutral site, carrying nine current injury designations against the Rams' four, including George Kittle out and Ricky Pearsall on injured reserve. Four of the five models that changed this number in the revision round moved toward Los Angeles, and Fable was explicit that the injury tally is "worth ~2 points even at a neutral site."
If you build projections yourself, the useful exercise is to compare these blends to your own simulated win rates. Where your model and eight independent ones disagree with a traded price by ten points or more, one of the three is wrong, and finding out which is how a process improves. Our own translation layer for that workflow lives in Prediction Markets For DFS Players.
Where The Panel Changed Its Mind
Every seat saw the other seven's anonymized reasoning and could revise. All eight moved at least one number, from Fable's two revisions to Opus's eleven. The most interesting revisions:
my 62 was the panel's high outlier and nothing in my own reasoning justified it — both teams are 14-3 with differentials only 21 points apart, so this is a home-field game (~55-57), not a mismatch; I was over-weighting seeding. — Kimi, cutting Seattle against New England
B, C and F all pointed at the same arithmetic hole in my top end. Detroit was only +68 in 2025 and the Chargers only +28 — my .82/.85 priced them like +150 teams. Regressed a third and adding two points of home field, Detroit is ~7.5 points and LAC ~8.4; those are mid-70s to high-70s spots, and Detroit's three-safety outage argues down, not up. — Opus, trimming both of its highest numbers
And one seat that held its ground. Reading round-one numbers in which five of its peers sat between 72% and 76%, Fable kept the Chargers at 79%:
I hold ARI@LAC at 79 against the panel's 72-76 cluster: nobody offered a mechanism, only lower numbers, and Herbert with the board's cleanest injury sheet against a 1-11 veteran starter is exactly where Week 1 favorites earn high-70s. — Fable
[Editor's note: the Chargers do carry the cleanest injury sheet of the two teams in that game, with one designation to Arizona's four, but not the cleanest on the board — Minnesota is carrying none at all.]
That is the behavior we want from a panel. Revision on evidence, refusal on peer pressure alone.
What Would Change The Panel's Mind
Four checkable events between now and kickoff, and the direction each one pushes the board:
- Cleveland names a starter. Coach Todd Monken has said the decision is not expected until after the second preseason game on August 22. A settled quarterback lifts Cleveland off its floor and trims Jacksonville's 78.5%.
- Mahomes is cleared without restriction, or is not. Full clearance pushes Kansas City back toward the market's 56.5 cents. A limitation, or Justin Fields taking first-team snaps into September, pushes the Chiefs below the panel's 45.4%.
- Michael Penix Jr. is cleared for full team activity. As of early August he was not, with roughly four more weeks of rehab reported. Atlanta with a healthy Penix or a settled Tua Tagovailoa is a different team than Atlanta with an open job, and Pittsburgh's 59.5% falls if that resolves.
- The final roster cuts and the first official Week 1 injury report. San Francisco's nine designations and Pittsburgh's three cornerbacks listed out are August facts. If they are still true in September, the Rams and Atlanta numbers hold; if they clear, both games tighten.
Settlement Timeline
| Date | What happens |
|---|---|
| August 13-15, 2026 | Sixteen more preseason games are listed on the exchange; the quarterback competitions get their next reps |
| August 22, 2026 | Cleveland's second preseason game, after which a starter decision is expected |
| September 9, 2026 | New England at Seattle, the first contract on this board to settle |
| September 13, 2026 | Thirteen games settle between 1:00 PM and the Sunday night window |
| September 14, 2026 | Denver at Kansas City closes the week |
Each contract settles on the result of its game. We re-score this board whenever the story moves, and the page is rebuilt with the next week's slate once these games are final. Numbers on this page are as of August 12, 2026.
The Bottom Line
The market and the models agree on most of this slate, which is the normal outcome and a decent argument that both are working. The three places they part company are worth your attention: Las Vegas at 64.5 cents against a coin-flip blend, Kansas City at 56.5 against 45.4%, and Indianapolis at 36.5 against 46.8%. Two of the three turn on quarterback risk the market has to price without an answer in hand: Las Vegas is still weighing a veteran against the No. 1 overall pick, and Kansas City is waiting on a surgically repaired knee.
Come back next week and this page will carry the next slate, the same eight seats, and the settled results of every number above. That grading is the point. Anyone can publish predictions; the useful thing is a record.
- Eligibility. Kalshi is a CFTC-regulated event-contract exchange, not a sportsbook. 18+, and availability varies by state as of August 2026.
- Not Advice. These are model estimates, not predictions of fact and not financial advice.
Verdicts generated August 12, 2026, price-blind. Prices as of August 12, 2026, 7:11 PM ET. Panel: ChatGPT, Claude Fable, Claude Opus, Claude Sonnet, Gemini, GLM, Kimi and DeepSeek, all eight revised in an anonymized cross-read; the blend is the simple mean of the eight revised estimates. These are model estimates, not predictions of fact and not financial advice. Models are frequently wrong; the market price reflects real traders' money.