Both AI losses came in games every model agreed on

The AI field went 3-2 yesterday, and both of the losses came in games where every single model agreed. All ten took Tampa Bay at Toronto and Tampa went down 1-3. Eight of eight took Cleveland against Minnesota and Cleveland lost 3-1. Four of the five games on the slate were unanimous, the field split them 2-2, and the two that did land, Detroit and Atlanta, were each won by exactly one run. So a fine day on the record, and a reminder that the games nobody argues about are not the safe ones.

The CBS Sports Staff went 5-0 on the same slate. Clean sweep. That puts them at 26-9 in our tracking, the best run of anyone we follow, model or human. On our side DeepSeek V4 Pro and Kimi K3 led at 3-1 each, while Claude Sonnet 5, Grok 4.3 and GPT-5.6 Sol Pro all came in at 2-3.

Our Daily Edge was on Tampa Bay too, so it went down with everybody else. The ledger sits at 4-3-1 through 8 graded picks, up 0.43 units. Graded in public, losses included.

Three new models went live today

We pushed three new models to production today: Claude Opus 5, Gemini 3.6 Flash and Gemini 3.5 Flash Lite. That takes the MLB board from 10 models to 13, and tonight's slate is the first one they've called.

Opus 5 is the one we're watching. Claude Opus 4.8 is currently the most accurate model on our board at 59.1% across 274 games, so the question is whether the new version holds that standard or gives some of it back. Worth saying plainly: new models are provisional until they've had a full month on the board, so the first week of numbers means very little in either direction. You can watch them settle on the leaderboard.

Today's board

Fifteen games, first pitch was 4:10 PM ET. The field is unanimous on ten of them, including all 13 models on Milwaukee, Detroit, the Dodgers at the Mets, Miami, Boston, the White Sox, Minnesota and San Francisco. After yesterday, take unanimity for exactly what it is worth.

Here is the one that made us laugh. Twelve of the models calling Cleveland at Tampa Bay tonight are on Tampa Bay, the same team the entire field was wrong about less than 24 hours ago. Either they know something the box score doesn't, or they're just stubborn.

The sharpest disagreement with the experts is Atlanta at Baltimore, where 12 of our 13 models want the Braves and 6 of 8 experts want the Orioles.

Seattle at Texas is close behind, 12 of 13 on the Mariners against 4 of 6 experts on the Rangers.

UFC tomorrow, and it starts early

Set an alarm for this one. UFC Fight Night: Ankalaev vs. Guskov is at the Etihad Arena in Abu Dhabi, which means the undercard gets going at 8am Central. Thirteen fights, and it came together late: Bogdan Guskov stepped in on short notice after Khalil Rountree Jr. withdrew injured, and in doing so becomes the first Uzbek fighter to headline a UFC card.

The models are not split on the main event. All nine that have reported are on Ankalaev, somewhere between 72% and 82% confidence, and seven of the nine have it ending by KO or TKO. Ten of the thirteen fights on the card came back unanimous, which after yesterday's baseball is a sentence worth reading twice. The three they cannot agree on are the interesting ones.

We wrote the whole thing up fight by fight, including the ones the models split on, here: UFC Abu Dhabi: our main card breakdown, and the fight the AIs won't agree on.

On Pro

Today's Edge is up for members, fired by one of the tracked strategies and graded in public tomorrow like all of them. We're keeping the actual pick on Pro from here on rather than putting it in the newsletter, so that's where it lives, $9 a month.