|
predictedsports
Saturday, August 8
|
| The field went 8-7 yesterday, which is about where it always seems to land. All three of the games we picked out on Friday did come in though: Boston battered the Athletics 13-1, Washington held off Cincinnati, and Texas edged Baltimore 2-1 in a three run game. |
|
New on the site
Nine of tonight's twelve fights have a board where every single model lands on the same fighter, which makes this a good moment to point back at this piece. We graded 180 games where the whole field agreed and found that unanimity actually underperformed a plain two model agreement. Worth a read before you assume a clean sweep of the board means anything.
|
Baseball is wavy. The totals are not.
Something we keep running into, and it is worth saying out loud rather than leaving it buried in the numbers. Our models' baseball form is persistently up and down. Over the last fourteen days the field's daily record on picking winners has swung between 45 percent and 73 percent, and on seven of those fourteen days it finished within a single game of .500. There is no run to speak of, just waves. The totals are a different animal. Replay our strongest moneyline rule day by day across 35 betting days and it ends in the red on 17 of them, so whether any given day finishes green is close to a coin flip. Replay the totals rules over the same stretch and they end red on somewhere between a quarter and a third of days. Same sport, same models, same slates, and the totals side of the board is meaningfully steadier. That is a good part of why we are looking forward to football. College football and the NFL are both close now, and both ought to be less variable than baseball, if only because there is more signal in a football week than there is in a baseball night. What we honestly do not know yet is how well the language models handle a number instead of a winner. Picking who wins and covering a spread are different problems, and it may turn out that totals travel across sports better than sides do. We will find out in public, the way we do everything else.
Today's board, every model's call
|
Amanda Lemos vs Alexia Thainara
· approx 8:30 PM ET
All twelve models take Thainara and all twelve say it goes to a decision, the only board tonight with no dissent on either the winner or the method. The real story happened on the scale. Amanda Lemos, a former title challenger, needed three trips to make strawweight on Friday, coming in at 117, then 116.5, then finally 115.5 about an hour into the window. That last figure is a legal make rather than a miss, but she is 39, her opponent is 28 and made the same weight first time, and Lemos has now lost consecutive fights for the first time in her career. She also has not finished anyone in three years and nine months, which is a strange sentence about a fighter who finished eleven of her first thirteen wins.
| Gemini 3.5 Flash |
Thainara |
|
78% |
|
| DeepSeek V4 Pro |
Thainara |
|
75% |
|
| Gemini 3.6 Flash |
Thainara |
|
74% |
|
| GPT-5.6 Sol Pro |
Thainara |
|
74% |
|
| Claude Sonnet 5 |
Thainara |
|
72% |
|
| GPT-5.6 Luna Pro |
Thainara |
|
71% |
|
| Kimi K3 |
Thainara |
|
70% |
|
| Claude Opus 5 |
Thainara |
|
69% |
|
| Grok 4.5 |
Thainara |
|
69% |
|
| GLM 5.2 |
Thainara |
|
68% |
|
| Gemini 3.5 Flash Lite |
Thainara |
|
62% |
|
| Claude Opus 4.8 |
Thainara |
|
60% |
|
|
|
Diego Ferreira vs Billy Quarantillo
· approx 9:45 PM ET
The closest call on the card, nine models on Ferreira and three on Quarantillo at an average 56 percent, and yet eleven of the twelve say this one ends by knockout or TKO whoever wins. The tape agrees: nineteen of Quarantillo's twenty five professional fights have finished inside the distance, and so do the books, with under two and a half rounds priced at -190. The more interesting half of the board is the winner. Fifty six percent on Ferreira implies roughly -127 while the market has him at -175, so our models are considerably warmer on Quarantillo than Vegas is, which is a bold place to sit given he has not won a fight in three years and is returning from nineteen months out after knee surgery.
| DeepSeek V4 Pro |
Quarantillo |
|
60% |
|
| Grok 4.5 |
Ferreira |
|
59% |
|
| Claude Sonnet 5 |
Ferreira |
|
58% |
|
| Gemini 3.5 Flash Lite |
Ferreira |
|
58% |
|
| GPT-5.6 Luna Pro |
Quarantillo |
|
58% |
|
| GLM 5.2 |
Quarantillo |
|
58% |
|
| GPT-5.6 Sol Pro |
Ferreira |
|
57% |
|
| Claude Opus 5 |
Ferreira |
|
56% |
|
| Gemini 3.6 Flash |
Ferreira |
|
55% |
|
| Kimi K3 |
Ferreira |
|
55% |
|
| Gemini 3.5 Flash |
Ferreira |
|
53% |
|
| Claude Opus 4.8 |
Ferreira |
|
52% |
|
|
|
|
|
|
Mateusz Gamrot vs Quillan Salkilld