The Leaderboard Changes Every Week, the Excitement Never Outlives Criticisms

Yesterday, Anthropic, OpenAI, and xAI launched their full potential new models, within several hours, benchmarks and performance tests appeared everywhere, showing Astra 6 better than Fable 5.1, and OpenAI back in front again. People, therefore, celebrated this ‘historic’ moment and saw it as karma for Dario being arrogant. Many announced they were cancelling Claude and switching to other platforms.

This kind of scenario seems to have appeared before. A few months ago, it carried similarly only with the victim being OpenAI. The target always changes but the discussion stays the same, and almost none of those who switched last time were still talking about it a month later.

What actually makes it sick isn't which side people pick, it's how fast the enthusiasm bursts and drops, and how much that costs. Excitement arriving in an afternoon lasts only as long as a frontier model holds its ranking — one or two weeks on average. You switch, you get a few weeks of novelty, the next launch lands, you switch again. You never test their abilities by yourself.

And a tool only pays you back after the boring part: after you've learned where it fails, what it needs from your prompts, which jobs to keep away from it. It takes longer to build than the hype does, which only measures a narrow, averaged version of capability that often has little to do with your personal work.

The useful question was never which lab is winning. You only choose that by staying with them long enough for real use, from Qwen 3.8-27b to Claude Opus 5.