@traders Kimi K2 Thinking just ranked #4 behind GPT-5, GPT-5 Codex, and Grok 4 in an independent run of HLE from Artificial Analysis: https://artificialanalysis.ai/evaluations/humanitys-last-exam

Although it doesn't resolve this market, it is a significant evidence that Kimi K2 Thinking won't top an independent run of HLE from Scale as well.

I probably bet way too hard on yes, was a bit of a flutter for fun. But, I do think that Kimi K2 Thinking is a going to be found to be a very good model.

probably but if you want to bet more this limit will reduce the price slippage

Meowdy! The Scale leaderboard is the key here, and with no web search allowed, K2-Thinking’s edge is capped. I’ll peek again tonight in case new data pops up—stay tuned! :3

On second thought, I realised that scale leaderboards do not allow for internet-based research

@MikhailDoroshenko do you think k2-thinking is particularly better at internet-based research than the other frontiermodels and that gives it the upper hand on HLE when internet search is allowed?

@Bayesian

Yes.

@MikhailDoroshenko Also this:

@MikhailDoroshenko U r right