It might vary between tasks though. A model that’s great at abstract reasoning might be great at writing math proofs but struggle to write software in <insert language>.