Positive users praise Vercel AI Gateway for standardizing providers, handling cold starts, and reliable fallbacks, while some negative users criticize the need to reverse engineer APIs and stacked real-world cold starts.
No Digg Deeper questions have been answered for this story yet.
Vercel AI Gateway is the fastest way to get tokens, cold or hot, across the board. DMs open if you need help migrating or benchmarking your current solutions.
Full picture: My table showed cold connection setup (DNS+TCP+TLS). Real and reproduced. But setup ≠ end-to-end. Same model, same prompt, interleaved runs: @vercel wins TTFT from here, cold or warm. @cloudflare and @openrouter win the connection phase, which is what short-lived CLI/desktop processes feel most. Both things are true. Both statements are true, you may want to pick per workload. Open sourced my benchmark if you want to reproduce it from your region: https://github.com/rbadillap/ai-gateways-benchmark
How? AI Gateway runs on Vercel itself. It takes advantage of Fluid compute and our optimized global network (that never mis-routes to random "POPs"!) We wrote about how we did this. You can even build your very own fast gateways on Vercel: https://vercel.com/blog/how-ai-gateway-runs-on-fluid-compute
Vercel AI Gateway is the fastest way to get tokens, cold or hot, across the board. DMs open if you need help migrating or benchmarking your current solutions.
@rauchg i wonder if anyone will switch over from @OpenRouter not sure if there's any reason to rn
@rauchg Do you have plans to make EU data residency available in the near future?
@esnx_xyz @rauchg @OpenRouter doesn't openrouter charge a fee on top of the llm cost? vercel ai gateway is free
@llmgateway @rauchg @llmgateway what surprised you most in the comparison?
@rauchg Feedback, we had to reverse engineer the video model api from your SDKs which is not optimal.
@rauchg Any plans to compete or offer something like OpenCode Go ?
@rauchg I’ll test it via http://tokensize.dev
@rauchg stacking a Supabase edge function cold start on top of a gateway cold start is brutal. the benchmark assumes a warm caller. real prod number ended up almost double the advertised cold figure.
@0xMoe_ @esnx_xyz @rauchg @OpenRouter vercel is not free
@rauchg been routing our agent calls through it at work, the fallback behavior when a provider rate limits mid request is what actually sold me not the latency numbers
@rauchg why would you have "CF AI Gateway -> OpenRouter" instead of just CF AI GW by itself?