Teortaxes also claimed the model beats Claude Opus 4.8.
@TheZvi What "yep"? It does beat Opus 4.8 on some legit benchmarks. The Mythos stuff is of course ludicrous (though it does beat Fable on PostTrainBench but that's almost certainly due to Ant's obnoxious stealth sandbagging). But as far as public models go, it is in the same ballpark.
Yep.
Negative users criticize claims of DeepSeek topping Opus on benchmarks for misrepresenting the intended context and message.
No Digg Deeper questions have been answered for this story yet.
@teortaxesTex The context and intended message being sent are not the context and message you are suggesting. Even ignoring Mythos he is making a much stronger, and clearly false, claim.
@TheZvi In a narrow technical sense the screenshotted tweet is not clearly false. I'd say "no compromises" is false but Claude use case distribution is so lopsided towards coding now, it barely counts I've lost track of the narrative Sacks and pmarca are advancing at this point