Comparing LLM API Gateways — The Traps in Markup, Billing, Regions, and Training-Data Use

Bundling multiple LLM APIs (OpenAI, Anthropic, Google) behind one endpoint has become a standard component of any generative-AI app. Using Vercel AI Gateway as the reference point, this post compares OpenRouter, Cloudflare, Helicone, Portkey, and LiteLLM, plus Bedrock, Azure, and Vertex AI, along four axes: markup, billing model, region, and training-data use. The shared trap: zero markup does not mean zero cost.