Hermes · Provider Routing

Same model, many sub-providers. On OpenRouter, you pick the priority: cost, speed, or quality.

OpenRouter doesn't run models itself, it farms requests to many underlying providers (Anthropic, Google, Bedrock, Together). Provider routing controls which of them handle your requests and how they rank. The main knob is sort: price (cheapest), throughput (fastest tokens/sec), or latency (quickest first token). Then get specific with only (whitelist), ignore (blacklist), and order (preferred, then fall back). Two guards worth setting: require_parameters: true (no silent drops of tools/temperature) and data_collection: "deny" (prompts stay out of training). It applies only on OpenRouter, and it's sub-provider routing, not the same as fallback providers switching to a different provider on failure.