OpenRouter alternatives: 7 LLM gateways compared

OpenRouter is a hosted service that gives you “hundreds of AI models through a single API endpoint”, according to its quickstart. You can point the OpenAI SDK at it as a drop-in replacement, buy credits once and call models from many providers with one key.

What it does well, from its own documentation:

• Provider routing and fallback. By default, the provider routing docs say requests are load balanced across top providers, weighted toward cheaper ones, with the rest used as fallbacks. You can set a provider order or sort by price, throughput or latency.
• No markup on inference. Its pricing page says inference is billed at each provider’s list price on every plan.
• Data controls. The FAQ says prompts and completions are not logged by default, and routing can be limited to providers that do not collect data or that offer zero data retention.
• Spend controls. Its spend controls guide describes guardrail budgets per person or per key, and pooled workspace budgets.
• Provider prompt caching. It passes through prompt caching on supported providers and tries to send follow-up requests to the same provider to keep that cache warm (docs).

These reasons follow from how OpenRouter itself says it works, not from complaints:

• The credit-purchase fee adds up at volume. 5.5% or 8% on every top-up is small for a prototype and a real line item at large monthly spend. Some alternatives charge no fee on tokens, others are free software you run yourself.
• You already have provider contracts. If you hold your own OpenAI, Anthropic or Google agreements, committed-use discounts or cloud credits, a gateway that calls providers with your keys avoids the BYOK fee above the monthly allowance.
• Data must stay in your cloud. OpenRouter is hosted, so every prompt passes through its infrastructure on the way to the provider. A self-hosted gateway keeps that hop in your own network.
• You want a response cache. OpenRouter’s caching docs describe provider prompt caching, not a cache that answers repeated requests without calling a model. Several alternatives offer exact or semantic response caching.

Vercel AI Gateway: the closest hosted swap

Like OpenRouter, Vercel sells model access through credits, with provider ordering and model fallbacks. Its pricing docs say it “charges no markup and no platform fee on tokens”, and BYOK carries no fee either; you are responsible for payment processing fees, and some features such as custom reporting and team-wide zero data retention are paid add-ons. Budgets are soft caps and cover spend on Vercel’s credentials, not BYOK spend.

Cloudflare AI Gateway: caching and rate limits in front of providers

Cloudflare’s pricing page says the core features — analytics, caching and rate limiting — are free. You can use your own provider keys or buy credits through Unified Billing, which carries a 5% fee. It suits teams already on Cloudflare that want a response cache, which OpenRouter does not offer.

LiteLLM, Portkey, Bifrost: self-hosted, with your own provider keys

These run in your cloud and call providers with your API keys, so there is no fee on top of provider prices, but you operate the gateway. LiteLLM has the broadest open-source feature set; Portkey keeps semantic caching and granular budgets for its paid tiers; Bifrost lists budgets and semantic caching in its Apache 2.0 build. If you like LiteLLM but not the operations, Perimattic ManageStacks offers managed LiteLLM hosting. See LiteLLM alternatives for more on this group.

Agent Router and Kong: for platform teams

Agent Router (formerly Envoy AI Gateway) suits teams running Envoy on Kubernetes. Kong AI Gateway suits teams already on Kong, but its load-balancing, semantic cache and token rate-limiting plugins are AI Gateway Enterprise only.

• You want the most models with the least setup. One key, hundreds of models and nothing to run.
• You do not have provider contracts. OpenRouter handles accounts and billing with each provider for you, and bills inference at list price.
• You are experimenting. Trying many models quickly is easier through one catalog than through separate provider sign-ups.
• Your spend is modest. At low volume, the credit-purchase fee is likely smaller than the cost of running and maintaining your own gateway.
• Your data rules are met by its controls. No logging by default, data-collection filters and zero data retention routing may be enough for your policy. Check with your security team.

Perimattic LLM Gateway is left out of the comparison above, which covers third-party tools. It is built for teams that want one API across providers, using their own provider accounts, with cost controls built in. Its features:

• One OpenAI-compatible API in front of OpenAI, Anthropic, Google, Mistral, DeepSeek and open models.
• Routing rules by cost, latency or task, with automatic fallback when a provider errors or rate-limits.
• Exact-match and semantic caching of responses.
• Budgets and rate limits per team, API key or customer.
• Cost tracking and attribution by team, feature and customer, with alerts.
• Request logging with PII redaction.
• Runs in your own cloud (VPC) or managed.

Get started: Sign up and our team will get in touch, or read the LLM gateway overview.