It can be easily gamed. The users are self-selected, and they have zero incentive to be honest or rigorous or provide good responses. Some have incentives the opposite way. (There was a report of a prediction market user who said they had won a market on Gemini models by manipulating the votes; LMArena swore furiously there had definitely been no manipulation but was conspicuously silent on any details.) And the release of more LMArena responses has shown that a lot of the user ratings are blatantly wrong: either they're basically fraudulent, or LMArena's current users are people whose ratings you should be optimizing against because they are so ignorant, lazy, and superficial.

At this point, when I look at the output from my Gemini-2.5-pro sessions, they are so high quality, and take so long to read, and check, and have an informed opinion on, I just can't trust the slapdash approach of LMArena in assuming that careless driveby maybe-didn't-even-read-the-responses-ain't-no-one-got-time-for-that-nerd-shit ratings mean much of anything. There have been red flags in the past and I've been taking them ever less seriously even as one of many benchmarks since early last year, but maybe this is the biggest backfire yet. And it's only going to get worse. At this rate, without major overhaul, you should take being #1 on LMArena seriously as useful and important news - as a reason to not use a model.

It's past time for LMArena people to sit down and have some thorough reflection on whether it is still worth running at all, and at what point they are doing more harm than good. No benchmark lives forever and it is normal and healthy to shut them down at some point after having been saturated, but some manage to live a lot longer than they should have...