Source: OpenRouter provider routing and model fallbacks docs
OpenRouter's reference for how it load-balances one model across many inference providers, the provider object (order, only, sort, max_price, zdr...), and the models-array fallback.
Contents
- What: official documentation (“Provider Routing”, “Model Fallbacks”)
- Publisher: OpenRouter
- Date: living docs, read 2026-11-14
- Original: Provider routing, Model fallbacks
Summary
OpenRouter is a hosted aggregator: one model slug may be served by many providers (with different prices, quantizations and data policies). By default it picks a provider at random, weighted toward the best on a composite of price, throughput, latency, 429 rate and reliability, and falls back to other providers on errors. Clients can override this with a provider object and list fallback models.
Key claims
- Default routing weights traffic toward providers near the best on effective price, throughput, time to first token, 429 rate, reliability and tool-call quality, while still sending a small share to others so improvements are noticed.
providerfields:order,allow_fallbacks(default true),require_parameters(only providers supporting every request parameter),data_collection(allow/deny),zdr,only,ignore,quantizations,sort(price/throughput/latency; bypasses default weighting),max_price(hard ceiling),preferred_min_throughput/preferred_max_latency(soft, percentile-based).- Slug suffixes
:nitro(sort by throughput) and:floor(sort by price) are routing variants, not separate models; they also make priority (:nitro) or flex (:floor) service-tier endpoints eligible. Thepreferred_*thresholds use a rolling 5-minute window and only reorder endpoints, whereasmax_pricecan block a request. - Model fallbacks: a
modelsarray in priority order; by default any error triggers the next model, including context-length errors, moderation flags, rate limits and downtime. Billing is at the price of the model that answered, and the responsemodelfield shows which one did.