Skip to main content
MANIFOLD
Will a Chinese AI model rank in the top 10 on the LMArena Text leaderboard on August 1?
41
Ṁ100Ṁ3k
resolved Aug 2
Resolved
NO

This is one of 25 props in the 11th annual Narcissist Forecasting Contest, as described here:

https://braff.co/advice/f/announcing-the-2026-narcissist-forecasting-contest

The prop and any ambiguities will be resolved by a panel of human judges, as described in the fine print of the entry form:

https://forms.gle/nDDxdxyYSNheY9FV7

This means that the prop may resolve differently from how other, similar-looking Manifold props resolve. Do not bet on this market if that is going to bother you.

Market context
Get
Ṁ1,000
to start trading!

🏅 Top traders

#TraderTotal profit
1Ṁ285
2Ṁ73
3Ṁ52
4Ṁ48
5Ṁ45
Sort by:

style-control on or off?

@Eternal good question. It's on by default, which gives a reading of Kimi #12. Turning it off, top Chinese model is Qwen at #13. There's a separate filter called Factuality, under which Qwen is #10. But I suspect the Yes bets here are based on something else: reading the prop as "by August 1" instead of "on August 1."

@AdamBraff So you'd resolve based on default?

@JasonMendoza2008 if there's no live controversy (i.e., nobody seriously arguing the Yes side) and if the status quo remains through end of day, I will resolve this as No. But if there's a genuine dispute, it goes to the judges in my longstanding forecasting contest, of which this prop is a mirror.

https://braff.co/advice/f/announcing-the-2026-narcissist-forecasting-contest

Kimi now at #12. This seems pretty clearly headed to No. Can the Yes bettors explain their position so that I can put the argument to our judges? Thanks

Will wait until end of day tomorrow August 1 to resolve this. Not sure why it's trading at 90% given that Kimi-k3-max is the top-rated Chinese model and it's been sitting at #11 all week. Anyone want to explain the interest in Yes here, in case there is something I need to explain to the contest judges?

https://arena.ai/leaderboard/text

@AdamBraff I assume the interest in the yes side here is due to rank spread (do we take confidence intervals into account?) and/or grouping of identical models at different thinking effort settings (can the same model count multiple times in the top 10?).