Adi Reports Astra-Low Dominating Eyebench
18-year-old LLM engineer shares surprise at new model's benchmark results.
TLDR
Adonis Singh posted that astra-low beats every single model on eyebench by a wide margin, including 5.6-sol max. He stated he has no idea how to create a v4 version because the benchmark already proved absurdly difficult for models a few weeks earlier and for himself. Singh suggested it might be time to retire the bench soon. The post came from the 18-year-old engineer who probes LLMs, interned at LM Studio on inference, and contributes to MCBench.
Combined views
32.1K
1 Source, first seen 26d ago