CBS went 14-3. The machines went 8-9.

Every so often the humans remind the machines who taught them. Yesterday was one of those days. Across 17 games the AI consensus went 8-9, a losing night, while the CBS Sports staff went 14-3. The best of our tracked experts had one of the sharpest single days we have on record, and they did it while our field was underwater. Claude Opus 4.8 salvaged some pride at 10-7 to lead the models, Kimi K3 brought up the rear at 6-11.

It is a light one today, just 5 games with a 12:15 PM ET first pitch, and the field is unanimous on Atlanta, Cleveland and Tampa Bay. One thing about that field is different now, though.

Grok 4.3 is back, because 4.5 lost the plot

Here is a strange one. We track every model version by version, and Grok 4.5, the current release, has been quietly awful at run totals: a 42 percent hit rate and a 15 percent loss on the season. Its predecessor, Grok 4.3, was the opposite, one of the best totals forecasters we have ever tracked, at a 64 percent hit rate and a 25.8 percent return. Same lineage, one version apart, and the newer one is worse. The bump did not dull the edge, it inverted it. So we did the only sensible thing and brought Grok 4.3 back onto the board. It is still callable, it is already grading forward again, and it takes the field to ten models. We will keep both and let the record sort them out, in public, like everything else.

Our full UFC Abu Dhabi breakdown is up

Saturday the UFC lands in Abu Dhabi, and Magomed Ankalaev headlines his first fight since losing the light heavyweight belt, against short-notice replacement Bogdan Guskov. We ran the whole main card through our ratings, the tale of the tape, and our nine-model panel, and wrote the lot up. All nine models are on Ankalaev, but the fun is one weight class down: on Erceg versus Temirov, our model and the betting market both like Steve Erceg, and eight of our nine AIs took Ramazan Temirov instead. That is the kind of split we live for. The full main card, fight by fight, is right here.

New on Pro: every edge, charted

If you follow the winning strategies on Pro, there is something new to look at. Every edge now shows its full return chart, the running units-won line from its very first graded pick, so you can watch the climb (or the slump) instead of reading one season-long number. The Claude Consensus sits around plus 29 units, the Grok 4.3 Total around plus 26, the DeepSeek Under around plus 22, all flat-staked and graded in public, losses included. It is the honest version of a tout's screenshot: the whole line, not the good week. Come see them on Pro.

That is the board for today. Back tomorrow.