Claims Grok is better than other models in its compute class, by which it explicitly includes GPT-3. 5-turbo but excluded GPT-4. But how is getting any idea of “compute class”? Neither of the OpenAI models have official published information sufficient to judge this; 3.5 was rumored to be 175 billion parameters, though Microsoft's CodeFusion paper indicated that its a 20 billion parameter model. Lots of people throw around a 1.5 trillion parameter guess for GPT-4, but that doesn't seem to be grounded in anything but speculation (in part itself based on the 175 billion parameter estimate for gpt-3.5-turbo.)