For most serverless workloads in 2026, Cloudflare Workers is the cheapest, and it's not particularly close.
| Model | Requests | Cold starts | Bandwidth | |
|---|---|---|---|---|
| Cloudflare Workers | $5/mo base, 10M requests included | +$0.30/M | none (V8 isolates) | free |
| AWS Lambda | pay per use | 1M free, then $0.20/M | yes (microVMs) | egress charged |
| Vercel Functions | Pro $20/seat + compute | region-priced compute | some | included in plan limits |
Workers runs V8 isolates in 300+ locations with zero cold starts and zero bandwidth charges — $5/month covers 10 million requests, then $0.30 per additional million. For an API, an edge function, an AI agent backend, an MCP server, that pricing and that latency are hard to beat.
Lambda has the more generous free tier on paper (1M requests plus 400,000 GB-seconds of compute every month, permanently) and more raw power per invocation via Firecracker microVMs — but you pay for egress, you deal with cold starts, and the billing has more moving parts.
Vercel Functions only make sense in the context of a Vercel deployment — you're on Pro at $20/seat, and compute is priced by region ($0.128–0.221 per CPU-hour). Convenient if your app already lives there, expensive as a standalone compute layer.
Building an API or edge service as its own thing: Cloudflare Workers, for the price and the no-cold-start guarantee. Already deep in AWS with workloads that need real compute per call: Lambda, and lean on the permanent free allowance. Already deploying a Next.js app on Vercel: use its functions and don't add a fourth platform.