What it is
OpenRouter is a hosted model router. Your code calls one OpenAI-compatible endpoint with one API key, names a model such as anthropic/claude-sonnet or openai/gpt-5, and OpenRouter forwards the request to a provider that serves that model, then bills you from one prepaid credit balance. The pricing page lists 600+ models from 80+ providers. On top of the pass-through it adds provider routing (sort by price, throughput or latency, or pin and exclude providers), automatic fallbacks to a second model or provider when the first errors, per-key budgets, workspaces, and data-policy routing that keeps requests on providers with zero data retention.
The operational problem it solves: an ops team has agents in n8n, a Python script or two, and a tool like OpenClaw, each wired to a different vendor with its own key, invoice and rate limit. Finance cannot see spend per agent, and switching a workflow from one model to a cheaper one means a code change and a new contract. OpenRouter makes the model a string in the request and the bill a single line.
Ownership changed this quarter. OpenRouter raised a $113M Series B led by CapitalG at a $1.3B valuation in May 2026. Stripe agreed to buy it on 19 August 2026 and closed the deal on 10 November 2026. Terms were not disclosed; press reports range from $7B to $8B. OpenRouter says the name, product, existing integrations and roadmap continue, and that it stays model-neutral. Treat that as a stated intent, not a contract term.
Pricing reality
There is no subscription. Inference is billed at each provider’s list price on every plan, and OpenRouter takes its fee when you buy credits, not per request:
- Free: 25+ free models, 50 requests a day (free models are capped at 20 requests a minute; buying $10 of credits once lifts the daily cap to 1,000). No BYOK.
- Standard: 5.5% fee on credit purchases, no minimum, five workspaces. Budgets, admin controls and zero-data-retention routing are included.
- Business (added September 2026, self-serve): 8% fee. Adds EU and US in-region routing, workload identity federation, up to 1,000 workspaces and a signed DPA on request. No SSO and no SLA.
- Enterprise: contract pricing with fee discounts at volume, invoicing and purchase orders, SSO (SAML) and SCIM, contractual SLAs, managed policy enforcement, and a dedicated support team. Private Endpoints, launched 14 October 2026, is an Enterprise-only add-on that routes to your own fine-tuned or reserved deployments, up to 10 per organization by default. No list price is published.
Bring your own key (BYOK) on Standard and Business: the first $25,000 of list-price inference a month carries no fee, then OpenRouter deducts 5% of what the same request would have cost on OpenRouter. Your provider bills you for the tokens.
Real-world math: an ops team spending $20,000 a month on inference pays about $1,100 in fees on Standard or $1,600 on Business. The same team on BYOK pays nothing extra until it crosses $25,000; at $60,000 of BYOK usage the fee is 5% of $35,000, or $1,750. On Enterprise, the fee discount is the main thing to negotiate, because OpenRouter states it does not discount inference.
Best for
RevOps, legal-ops and support-ops engineers at 20-500-person companies running several agents across two or more model vendors, who want one key, one balance, per-agent budgets and model fallbacks without running a gateway themselves, and whose security team accepts a third party in the request path.
Skip it if a single vendor contract already covers all your usage, if procurement requires SSO but cannot sign an Enterprise deal, or if data must never leave your cloud account.
Versus the alternatives
LiteLLM is the self-hosted counterpart: an open-source proxy you run yourself, with virtual keys, budgets and fallbacks and no per-dollar fee. Pick it when traffic must stay inside your network or fees at your volume exceed what it costs an engineer to run the proxy. Read its page first; it had two supply-chain compromises in 2026, so pin images and verify signatures.
Amazon Bedrock and Microsoft Foundry are the cloud-native route. Pick one when your company already has committed spend with AWS or Azure, needs data inside that cloud account, and can live with that cloud’s narrower model catalog.
A direct vendor contract with Anthropic, OpenAI or Google wins when more than about 80% of tokens go to one vendor: you skip the 5.5% fee and negotiate volume discounts on inference itself, which OpenRouter cannot give you.
Portkey, Vercel AI Gateway and Cloudflare AI Gateway are the other managed gateways to shortlist. Trial them if the Stripe ownership is a problem for your security review.
The rule: start on OpenRouter Standard while you are still choosing models. Move to BYOK once one vendor’s spend is steady, and to Enterprise or LiteLLM when you need SSO, an SLA or in-network traffic. Log every call to Langfuse or LangSmith whichever you pick, so the switch later is a config change. The AI agent ops stack shows where the gateway sits.
Watch-outs
- Payments-company ownership. Stripe now owns the layer that sees your prompts and your model spend. Guard: before your next renewal, get written confirmation of OpenRouter’s current data policy, turn on zero-data-retention routing for any workflow with customer or contract data, and keep your code OpenAI-compatible so you can repoint it to LiteLLM or a vendor in a day.
- The same model differs by provider. Providers can serve a model at different quantization, context length and price, and the router may pick a cheaper one. Guard: for production agents, pin an allow-list of providers per model and set the routing preference explicitly; do not rely on the default.
- “EU routing” covers inference, not records. EU in-region routing (Business and above) keeps requests on EU, UK or Swiss endpoints, but OpenRouter states account and billing records may be stored or processed in the US. Guard: have legal review the DPA against that sentence before you route personal data, and send EU traffic to
eu.openrouter.aiso requests fail rather than leave the region. - Fees scale with spend. At $100,000 a month on Business, the 8% fee is $8,000 a month. Guard: review the fee line every quarter and move steady single-vendor traffic to BYOK or a direct contract once it crosses the $25,000 allowance.