manual-reuse-d141/Research LLM benchmark performance over time, the rate of improvement, and project how long until SOTA frontier model capability is runnable on a phone (e.g. a 4B parameter model). Use empirical bench

anthropic_claude-opus-4-7_search_prxhub