“Modal has done the most to reduce cold-start times out of all the serverless inference solutions I’ve tried. They’ve put a ton of their engineering effort into system optimizations such as lazily loading container filesystems, and they give developers fine-grained control to trade off latency vs. cost.”