Stanford and Nvidia's open-source CLM-8B model matches Jev's performance while cutting agent decision latency to 16.5 ms, ...