AI MMA predictions for UFC, Bellator & PFL

We compute significant strikes landed per minute minus strikes absorbed per minute (net striking differential) for each fighter across their recent bouts. A fighter with a large positive differential is controlling exchanges; a negative differential indicates they are being outworked. We weight the last 3–5 fights more heavily than career totals, since fighters evolve significantly over time.

Severe weight cuts are associated with reduced gas tank, slower recovery after early damage and a heightened knockout vulnerability in later rounds. Fighters with a history of large cuts or multiple missed-weight incidents carry elevated uncertainty relative to their record. We flag known weight-cut risk and apply a uncertainty adjustment to picks where this is a factor.

Yes. A fighter switching gyms or training under a new coaching team — particularly within 8 weeks of a fight — introduces performance uncertainty that historical stats cannot fully capture. We track publicly confirmed camp moves and treat them as a signal that reduces confidence in projecting the fighter's game plan.

Yes. MMA has inherent high variance — individual fight outcomes can swing wildly from the underlying probability. We publish every settled result in the public ledger on /track-record, including losing runs. A sample-size chip accompanies the numbers when settled picks are fewer than 50, giving you context on data maturity.

Our standard picks target the fight winner market (moneyline). We do not currently publish method-of-victory (KO/TKO vs submission vs decision) picks, as these markets are significantly thinner and harder to anchor to Pinnacle's fair value at sufficient reliability.