Will any AI model reach a 1520 score on LMArena this year?

Anthropic’s September 1 release of Claude Fable 5.1 shot straight to the top of the LMArena leaderboard with a score of 1,514, putting it just two points ahead of Claude Opus 5 High at 1,505 and giving Anthropic six of the top 10 slots. One of the markets that prices this is Kalshi’s year-end cap contract market on how high the leaderboard’s top score will climb before January 1, 2027. With the bar already at 1,514, how much further can it realistically move in the remaining weeks?

The market currently prices “Yes” on at least 1,520 at 65 cents and “Yes” on at least 1,530 at 41 cents. We think the former is undervalued, and here's why.

Over the past 12 months, the top score has risen from about 1,475 to 1,514, roughly 40 points, or 3.3 points per month. Over the last six months, it has moved from about 1,503 to 1,514, just 11 points in half a year. Despite the slowing pace, extrapolating both the 6-month and 12-month rates to December 31 would still result in 1,521 to 1,527, which is above the 1,520 contract.

Another reason we believe the 65-cent price is undervalued is that LMArena adds ±11 near Claude Fable 5.1, resulting in a margin of error of 1,503 to 1,525. Not to mention that there are still 4 months left. That would imply 2-3 new models from each AI lab by the end of the year. Surely, the latest Anthropic models in December would be at least 6 points higher than Claude Fable 5.1, if not higher, even without considering the other labs' potential.