Comment by theptip

6 days ago

Are we reading the same chart? They have Sonnet <= high as Pareto dominant on $/perf.

You have to test each task obviously but it is not a bad model on its face.

They have updated it

  • Ha! So we were not looking at the same chart. That makes more sense.

    > Anthropic did post an official explanation, stating the original chart used a "simpler methodology" that "underestimated Sonnet 5's performance." The new chart supposedly uses their "standard methodology."

    Oops!

  • Did Anthropic have Opus 4.8 and Sonnet 5 switched in the Agentic Search chart at first?