alt Hacker NewsI dare anyone to convince me the benchmarks are not meaningless.
Wdym Opus 5.5 scores 14.7% higher than GPT Astra for Terminal Bench 4.0?
How would this alleged difference (most likely bs) actually show up in reality?
GPT Astra was literally the best model in the world by a margin until 1 hour ago or so.
Ah, so you didn't read the article.
>> On our benchmarks, Claude Opus 5.5 leads in agentic coding, computer use, and knowledge work. That said, at these levels of capability we’ve found that benchmark margins have become a less reliable guide to real-world differences. In our own use, the gap between Opus 5.5 and Claude Fable 5.1 is narrower than these scores suggest.