Mercury 2.5 LLM hits 770 tokens per second

The breakthrough in Mercury 2.5 demonstrates that inference speeds are reaching parity with human reading capacity, drastically lowering the latency overhead for real-time AI applications. This advancement shifts the competitive landscape toward extreme efficiency, proving that optimization is as critical as parameter scaling for model adoption.