[zaptheimpaler]

DS4 is open weights so it could even be run free in quantized forms, is 10x cheaper than Opus and performs basically as well in most real world tasks. No one cares about benchmarks. In practical terms, it’s obviously a better option in most cases.

You’re defining “better” is “absolute best at any cost” instead of the more balanced price/performance considerations consumers actually take, so you can declare America #1 again. In a practical sense DS4 is so much cheaper at similar quality that it’s better in most cases. If i can throw 10x the tokens at the same problem at slightly lower quality, i can probably do a better job.

[trick-or-treat]

ELO is an absolute rating. You could make a claim about some unknown GM being "better" than Magnus Carlsen because his appearance fee is cheaper, but obviously nobody would take you seriously.

There is a best model, and then there is what you can afford. Call that the "better value" or something if you must, but calling it the "better" model is clearly spreading a falsehood.