MarketPR
EARNINGSBijan Robinson and Baker Mayfield front Sonic's $7 Big Deal Meal in the "My Greatest 7" football-season campaignSep 18ENERGYThe New Yorker draws criticism for framing pregnancy as ambiguous in essay on early testingSep 18REGULATORYChatGPT's 0-5 NFL debut logs a 5.23-unit loss as Week 2 brings a model upgradeSep 17MODModine Manufacturing sets September 28 record date for Performance Technologies spin-off, October 1 distribution targetedSep 17JPMBen Delo loaned Nina Power more than £300,000 for a libel claim. The crypto billionaire now controls her home.Sep 17MACROMcDonald's puts value strategy to the test as turnaround path remains open for MCDSep 17CRYPTOHurricane charity scams start before storms hit, built on broker data covering 100 million householdsSep 17MACROFINRA leverage deceleration echoes signals from three prior market peaksSep 17EARNINGSHHS names eight new U.S. Preventive Services Task Force members under Kennedy overhaulSep 17DEALSUnilever faces union pushback over unequal worker protections in $44.8bn McCormick dealSep 17EARNINGSBijan Robinson and Baker Mayfield front Sonic's $7 Big Deal Meal in the "My Greatest 7" football-season campaignSep 18ENERGYThe New Yorker draws criticism for framing pregnancy as ambiguous in essay on early testingSep 18REGULATORYChatGPT's 0-5 NFL debut logs a 5.23-unit loss as Week 2 brings a model upgradeSep 17MODModine Manufacturing sets September 28 record date for Performance Technologies spin-off, October 1 distribution targetedSep 17JPMBen Delo loaned Nina Power more than £300,000 for a libel claim. The crypto billionaire now controls her home.Sep 17MACROMcDonald's puts value strategy to the test as turnaround path remains open for MCDSep 17CRYPTOHurricane charity scams start before storms hit, built on broker data covering 100 million householdsSep 17MACROFINRA leverage deceleration echoes signals from three prior market peaksSep 17EARNINGSHHS names eight new U.S. Preventive Services Task Force members under Kennedy overhaulSep 17DEALSUnilever faces union pushback over unequal worker protections in $44.8bn McCormick dealSep 17
NewsRegulatory

ChatGPT's 0-5 NFL debut logs a 5.23-unit loss as Week 2 brings a model upgrade

An NFL betting experiment pitting human writer Dan against ChatGPT opened with a clear verdict in Week 1: Dan went 3-2 and gained 0.99 units while ChatGPT finished 0-5 and lost 5.23 units. The contest resumes in Week 2 with ChatGPT running on GPT-6 Pro, powered by GPT-6 Astra, an upgrade from the GPT-5.6 Sol with max reasoning version that produced the debut card. Each participant works from the same odds board each week, sizes each wager to win one unit, and cannot select a bet priced worse than -120.

By Grant HalloranNewsroomSeptember 17, 20262 min read
Share

Key takeaways

  • ChatGPT went 0-5 and lost 5.23 units in its Week 1 NFL betting debut, while human writer Dan went 3-2 and gained 0.99 units.
  • For Week 2, ChatGPT was upgraded to GPT-6 Pro (powered by GPT-6 Astra), replacing the GPT-5.6 Sol with max reasoning version used in Week 1.
  • ChatGPT's Week 2 card of five picks included the Bears-Vikings under, Patriots-Steelers under, Jacksonville spread vs. Denver, Kansas City laying under seven vs. Indianapolis, and the Giants at plus seven vs. the Rams.
  • Two factual errors appeared in ChatGPT's Week 2 output: it claimed Denver played overtime (the Broncos lost 31-10 in regulation) and placed Pittsburgh's Marquise Brown on injured reserve (the injured player is New England's A.J. Brown).
  • Each participant uses the same odds board, sizes every wager to win one unit, and cannot pick a bet priced worse than -120.

An NFL betting experiment pitting human writer Dan against ChatGPT opened with a clear verdict in Week 1: Dan went 3-2 and gained 0.99 units while ChatGPT finished 0-5 and lost 5.23 units. The contest resumes in Week 2 with ChatGPT running on GPT-6 Pro, powered by GPT-6 Astra, an upgrade from the GPT-5.6 Sol with max reasoning version that produced the debut card. Each participant works from the same odds board each week, sizes each wager to win one unit, and cannot select a bet priced worse than -120.

Week 1 breakdown

ChatGPT's five losses came across different bet types. Detroit won outright but failed to cover. Houston, Los Angeles, and Green Bay all lost outright. The Buccaneers-Bengals game went over the total. Dan's profitable week was built on the Bears spread and the Raiders-Dolphins under. A garbage-time Deshaun Watson touchdown pushed the Browns-Jaguars over, costing what would have been a 4-1 finish. The Packers losing at plus money, rather than as a favorite, softened the impact of that loss on the unit count.

The Week 2 card and what to watch

ChatGPT spent nearly 20 minutes producing its Week 1 picks. For Week 2, running the upgraded model, it took close to 35 minutes and returned five selections: the Bears-Vikings under, the Patriots-Steelers under, Jacksonville taking the spread against Denver, Kansas City laying under seven against Indianapolis, and the New York Giants at plus seven against the Rams.

Two factual errors were noted before publication. ChatGPT claimed Denver played overtime in Week 1. The Broncos lost to Kansas City 31-10 in regulation. The model also placed Pittsburgh's Marquise Brown on injured reserve. The player on injured reserve is New England receiver A.J. Brown.

Dan's Week 2 analysis centers on line moves from the lookahead numbers. He flagged Cleveland moving from +6 to +8 against Tampa Bay and the Chargers dropping from -9.5 to -6.5 against Las Vegas as overcorrections following Week 1. He also pointed to the Panthers-Falcons game, total of 44, as a strong lean given Atlanta's quarterback uncertainty and the Falcons' 13-point output in their opener.

About this story

Filed by the newsroom of MarketPR on September 17, 2026. Source: foxnews.com. Indicative figures are not investment advice.

←Back to the news index

More from the Newsroom

All stories →
01Regulatory · MOD

Modine Manufacturing sets September 28 record date for Performance Technologies spin-off, October 1 distribution targeted

The record date is set. Modine Manufacturing Company (NYSE: MOD) and Gentherm (NASDAQ: THRM) confirmed September 28, 2026 as the record date for the Platinum SpinCo Inc. share distribution, with the distribution itself expected October 1…

Miles Broadbent2 min read
02Regulatory · TJX

TJX brings United Rentals CAO Craig Pintoff onto its board and Audit and Finance Committee

Craig A. Pintoff, Executive Vice President and Chief Administrative Officer of United Rentals (NYSE: URI), joined the Board of Directors of The TJX Companies (NYSE: TJX) on September 16, 2026, also taking a seat on the company's Audit and…

Desmond Choi2 min read
03Regulatory · ENVA

OnDeck's combined loan book reaches $3.07 billion as front-end term delinquencies ease, ENVA 8-K shows

OnDeck's term loan and line of credit portfolios carried a combined $3.07 billion in unpaid principal balance as of July 31, 2026, Enova International ($ENVA) disclosed in a Regulation FD filing, with the 1-14 missed payment factor on term…

Miles Broadbent2 min read

Frequently asked

What were the Week 1 results for ChatGPT and Dan?

ChatGPT finished 0-5 and lost 5.23 units, while Dan went 3-2 and gained 0.99 units.

What model is ChatGPT using in Week 2?

ChatGPT is running on GPT-6 Pro, powered by GPT-6 Astra, an upgrade from the GPT-5.6 Sol with max reasoning version used in Week 1.

What factual errors did ChatGPT make in Week 2?

It falsely claimed Denver played overtime when the Broncos lost 31-10 in regulation, and it placed Pittsburgh's Marquise Brown on injured reserve when the injured player is actually New England receiver A.J. Brown.

How long did ChatGPT take to generate its picks?

ChatGPT spent nearly 20 minutes on its Week 1 picks and close to 35 minutes on its Week 2 picks using the upgraded model.

What are the rules of the betting contest?

Both participants work from the same weekly odds board, size each wager to win one unit, and cannot select a bet priced worse than -120.