Beyond picking a model, it handles provider routing and fallbacks, enforces JSON Schema structured outputs, adds web search to any model, lets you bring your own provider keys, returns token and cost accounting in every response, and now runs server tools such as web search, an advisor model and a worker subagent. Configuration can live in dashboard presets instead of code. Inference is passed through at provider prices, with a platform fee on credit purchases and free-model caps on the free tier.
*Field guide compiled 2026-10-23 from the official documentation, pricing pages and changelogs linked below; every line links to the page that states it.*
What it does
- Presets. Named configurations, managed in the web app, that hold provider routing preferences, model selection with fallbacks, system prompts, generation parameters, provider include/exclude rules and tools, so you can change LLM setup without redeploying code. How: Create one at openrouter.ai/settings/presets, then send "model": "@preset/name" in the API request. source
- Provider Routing. Per-request control over which providers serve a model: order, allowfallbacks (default true), requireparameters, datacollection, zdr, only, ignore, quantizations, sort by price, throughput or latency, performance thresholds and maxprice. How: Add a provider object to the request, e.g. "provider": {"order": "anthropic", "openai", "allowfallbacks": false}. source
- Model Fallbacks. Automatically tries the next model if the primary model's providers are down, rate-limited or refuse for moderation; the request is priced by the model actually used, returned in the response's model field. How: Pass a models array in priority order instead of a single model. source
- Structured Outputs. Forces compatible models to return JSON that follows a given JSON Schema, avoiding parse errors and hallucinated fields. How: Set responseformat to {"type": "jsonschema", "jsonschema": {"name": ..., "strict": true, "schema": {...}}}. source
- Web Search. The web plugin adds search results to any model: native search for Anthropic, Google, OpenAI, Perplexity and SpaceXAI models, Exa for others, with engine choice, maxresults (default 5) and domain filters; citations follow a standard annotation schema. How: Append :online to the model slug, or send "plugins": {"id": "web"}; options include engine, maxresults, includedomains and excludedomains. source
- openrouter:websearch. Server-tool version of web search that lets the model decide when and how often to search instead of searching once per request; OpenRouter recommends it for better quality. How: Include openrouter:websearch in the request's tools. source
- BYOK. Use your own provider API keys through OpenRouter; by default your key is tried first and OpenRouter shared capacity is the fallback, with multiple keys per provider tried in priority order. How: Add keys under settings/integrations for each provider and choose prioritized, fallback or never-use-shared-capacity behavior. source
- Usage Accounting. Every response includes a usage object with native-tokenizer prompt and completion tokens, cost in credits, reasoning tokens and cached tokens, in the last SSE message when streaming, with no extra API call. How: Nothing to enable; read the usage field. The old usage: {include: true} parameter is deprecated and has no effect. source
- Advisor. Beta server tool openrouter:advisor lets a model consult any stronger OpenRouter model mid-generation at a decision point and continue with its guidance. How: Add the openrouter:advisor server tool to the request's tools. source
- Subagent. Beta server tool openrouter:subagent lets a model delegate self-contained tasks to a smaller, cheaper worker model, optionally with its own tools such as openrouter:websearch; the worker sees only the task description. How: Add the openrouter:subagent server tool; the model calls it with taskname and taskdescription. source
Pricing and limits
- Platform fee, Standard (pay-as-you-go): 5.5%. Rate limits passthrough from the provider(s); 500+ models, 80+ providers source
- Platform fee, Business: 8%. Rate limits passthrough from the provider(s); EU and US in-region routing; workspace limit 1,000 source
- BYOK: $25,000 of list price inference / month with no fees, 5% fee after. Standard and Business plans; Enterprise custom source
- Exa web search (web plugin): $0.007 per request. Includes up to 10 results; additional results $0.001 each, plus LLM usage for search result prompt tokens source
Traps
- If a :free model is followed by a paid model in models and OpenRouter falls back to it, you are billed at the paid model's price; list only :free models to keep a request free. source
- Web search incurs extra costs even with free models. source
- Google native search does not support domain filtering: with the default engine OpenRouter falls back to Exa when filters are set, and with "engine": "native" it returns a 400 error. source
- Server tools (advisor, subagent) are in beta and the API and behavior may change. source