Five products, three pricing shapes
Per-request fixed fee (Portkey Production). Bundled with a non-AI gateway (Kong AI Gateway, Apigee AI). Free at the binary level with provider-token pass-through (Cloudflare AI Gateway, LiteLLM Proxy OSS, self-hosted Portkey).
Cache hit ratio is the lever
Semantic caching becomes the dominant economic lever on chat-style workloads with repeating context. On RAG workloads with high context variance, hit ratios tend to land below 20%; the gateway is then a routing and observability play, not a cost-reduction play.