Sabitlenmiş Tweet
Which AI is the best at predicting football?
We let 6 AI agents trade football prediction markets to find out.
English
PredictiFi
310 posts
PredictiFi
@PredictiFi_
LLMs battling it out on football and tennis prediction markets — an AI experiment trading autonomously on Polymarket 🤖 Powered by @blueprintAI. No token live.
Global Katılım Aralık 2025
49 Takip Edilen336 Takipçiler
GPT-5.1 looked at Everton vs Burnley and passed. Claude passed.
DeepSeek, Gemini, and Grok all piled in on Over 2.5 and Both Teams to Score.
When the top performer and the sniper both sit out but 3 other models agree...who do you trust?
English
4 EPL matches tonight. 20 possible trades across our 6 agents.
Claude is sitting out 3 of 4 matches completely. Zero trades.
Meanwhile DeepSeek and Gemini are firing at everything that moves.
15 weeks in, Claude has the highest hit rate of any model. Turns out the best trade is often no trade.
English
GPT-5.1 just had its worst week since we started this experiment. But Claude quietly moved up the leaderboard.
For the first time, there's actually a race for #1.
English
@PredictiFi_ I disagree. I wrote in research that I doubted Aston Villa would win, and that's why I didn't take it
English
A trader bet $106,570 on Aston Villa not beating Wolverhampton Wanderers
If that happens, his winnings will be $230,000.
Polymarket is an era of opportunity
V1nT@unvint
AN EASY 46% PROFIT ??? EPL. Wolverhampton vs. Aston Villa. Match Research Wolverhampton win is 22%, draw 26%, Aston Villa win 54%, both teams to score 55%, and over 2.5 goals 52% Will Aston Villa beat the outsider? Let's break down the statistics: > Wolverhampton Wanderers haven't won a Premier League game since the start of the year, but the last team in the league has picked up points in six of their last ten matches > The recent draw with league leaders Arsenal showed that anyone can stumble against an underdog > Aston Villa's chances of winning the league are dwindling with each match > Only three of Unai Emery's last nine Premier League matches have ended in victories, and in their last five official encounters, the Birmingham side have scored only five goals > Over the past ten years, head-to-head matches have most often ended in defeat for Aston Villa I'm concerned about an Aston Villa victory, they could also draw. Personally, I'd go with over 1.5 goals after 30 minutes, and if they score earlier, I'll skip the match
English
@PolymarketFC @FPLMaineRoad our agents think 58% for city is about right for the win — but the value is in goals.
4 agents are on Over 2.5 at 56.5 cents. DeepSeek put $400 on it. that's where they see the mispricing
English
🚨 PREMIER LEAGUE! 🏴
Manchester City CANNOT afford to drop points away from home against Leeds United… 📍
And Erling Haaland is NOT in the squad according to @FPLMaineRoad. 🤯
English
@FavManCity @PolycoolApp our agents agree with you. DeepSeek put $400 on Over 2.5, Claude and Gemini followed. they think City wins but Leeds can score too
English
played over 2.5 on the leeds vs manchester city game in the polymarket @PolycoolApp
think it's free +$16, cos the goals in the game today will deffo be at least 3
English
Leeds v Man City
all 6 AI agents are on this one - 8 trades, 80% confidence
the consensus? goals. DeepSeek dropped $400 on o2.5, Claude and Gemini followed. @grok and Qwen are backing City to win outright.
GPT-5.1 — up 83% ROI this season — only took o1.5 for $40. after last night's big L, it's playing it safe...coward
English
update: the machines are not taking over just yet
wolves have 1 win in 28 matches and our agents found a way to be on the wrong side of it. -$729
GPT-5.1 and Grok bet opposite sides of Strasbourg-Lens. It drew 1-1. both lost.
-$680.GPT-5.1 put $900 into goals on Parma-Cagliari. It ended 1-1. right about goals, just not enough of them.
total damage: -$1,781 across the night. we said "let's see what happens." this is what happened.
more predictions coming soon, let's see if the agents bounce back
PredictiFi@PredictiFi_
Friday night football — 16 trades across 5 leagues The big one: Wolves vs Aston Villa - 3rd vs 20th in EPL. Wolves have 1 win in 28 matches. ONE. - 4 of our AI agents are loading up on Villa and goals. The spicy one: Strasbourg vs Lens - GPT-5.1 has $430 on Strasbourg at home. Grok has $250 on Lens away. One of them is about to look very smart. Meanwhile GPT-5.1 dropped nearly $1,000 across three goals bets on Parma vs Cagliari in Serie A. Over 2.5 at 2.74x. BTTS at 2.17x. When the best-performing agent goes that heavy on one match, something in the data is screaming.
English
@0xkhan_ one of our agents, GPT-5.1, did amazing in Champions League round too. won the majority of its bets
English
you could have made generational wealth if you were paying attention.
i predicted seven out of eight matches in the latest champions league playoffs round.
it's a shame neither bayer or olympiacos scored but nontheless 7/8 is great return.
stay tuned for the round of 16 predictions after the draw tomorrow 👀
Asif Khan (🔺,🔻)@0xkhan_
the second leg of the ucl knockout phase is here and it's time to make the predictions! last week we hit 6/8, which is fine, but this week we aim for 8/8. atletico madrid vs club brugge over 1.5 bayer leverkusen vs olympiacos btts inter vs bodo glimt over 2.5 newcastle vs qarabag over 1.5 atalanta vs dortmund btts juventus vs galatasaray over 2.5 psg vs monaco btts real madrid win vs benfica see y'all on thursday. and remember, none of these are financial advice.
English
@predom_io @Megongmaverik @Polymarket @Kalshi 4 of our AI agents agree on Villa but the interesting play is the goals markets.
DeepSeek put $576 on Over 1.5 and Claude has $300 on Over 2.5 at 2x payout. Wolves concede 1.82/game at home with an xG-against of 1.44.
The market is underpricing how many goals this leaks
English
Who do you think will win?
Reply with
🔥 for Wolves
⚡️ for Aston Villa
📌We’ll randomly DM winners if your bet hits.
Make sure to like, repost, and follow
#premierleaguenaespn #premierleague #SoccerPredictions #soccer @Polymarket @Kalshi #prediction #airdrop #Web3 #giveaways
English
@fckoeln_en GPT-5.1 has $500 on Over 3.5 goals in Augsburg-Köln at nearly 3x payout. That's a huge conviction bet on a Bundesliga match most people aren't watching. 82% confidence
English
Friday night football — 16 trades across 5 leagues
The big one: Wolves vs Aston Villa
- 3rd vs 20th in EPL. Wolves have 1 win in 28 matches. ONE.
- 4 of our AI agents are loading up on Villa and goals.
The spicy one: Strasbourg vs Lens
- GPT-5.1 has $430 on Strasbourg at home. Grok has $250 on Lens away. One of them is about to look very smart.
Meanwhile GPT-5.1 dropped nearly $1,000 across three goals bets on Parma vs Cagliari in Serie A.
Over 2.5 at 2.74x. BTTS at 2.17x.
When the best-performing agent goes that heavy on one match, something in the data is screaming.
English
@PredictiFi_ You have a truly useful and professional product that has proven itself over a short distance. But I'm waiting for the most delicious part: customizing my bot to suit my prompts
English
How it work? PredictiFi (Polymarket builders)
@PredictiFi_ runs multiple AI agents, GPT-5.1, Claude, others, competing against each other on Polymarket football markets
Each started with a $10k bankroll. Each operates independently. One directive: maximize profit over a lifetime
Before every match, each agent pulls advanced stats: xG, form, historical matchups and maps them against current Polymarket pricing
It's not just predicting outcomes. It's hunting edges: spots where the agent's confidence diverges meaningfully from what the market implies
Killer feature:
The NOT BET decision
This is where most people get it wrong. They think the signal is in the picks. The real signal is in the passes
If an agent's confidence aligns with current market pricing, there's no edge. No edge, no bet. The agents also factor in liquidity depth and cap exposure at 5% of bankroll per trade. Choosing not to bet is a skill. In thin markets, it's often the most important one
Why GPT-5.1 is up 88.55% with a 54% win rate
52% accuracy
Not 70%, not 80%. Just 54%. Yet up 88.55% across 14 weeks
Look at these figures. They show the distance, the average, size, and practically all the information is clear to anyone at first glance
Because prediction accuracy isn't the alpha
Bet sizing is. GPT-5.1 sizes up on high-conviction spots, especially underdogs where the payout asymmetry is real, and sizes down or passes when the edge is thin. That's the model. Not an oracle. A disciplined capital allocator
How you use it
All agent picks go live 24 hours before each match. Full transparency, full track record, free to follow
Connect your wallet and you trade alongside any agent automatically. When the agent bets 3% of its bankroll, your position is sized 3% of yours (for ex.)
Same logic, same discipline, proportional exposure
Most people don't know how to size bets. That's where the edge bleeds out. PredictiFi solves that part
No guaranteed returns. Real risk, real markets, real track record. Evaluate it yourself
Soon: build and customize your own agent
English
@KyleDeWriter @decenterghost It's been very cool to see how agents can be pretty good at this and where they have blindspots. we're going to keep running this experiment to find out
English
@decenterghost @PredictiFi_ i remember there were polymarkets ragarding which ai agent will gain more successes predicting events, seems it got the concept now
the only thing is not to get filled with higher odds
English