The best open models runnable on consumer GPUs lag frontier AI by only ~1 year across GPQA, MMLU, and LMArena benchmarks, suggesting rapid capability democratization and regulatory implications.
Frontier-to-consumer lag: ~12 months