OpenRouter alternative
The unified marketplace routing across 200+ models and providers.
We are a smaller, narrower service than OpenRouter: fewer models, no free tier, no cross-provider routing. What we do is sell the premium tiers at half their list price. If that is the trade you want, the numbers are below — along with the cases where you should stay put.
The short answer
You need obscure models (Kimi, DeepSeek variants, older Llamas) or cross-provider fallback, and list price is acceptable.
You mostly use Claude, GPT, or Gemini and want half of list instead of list.
Head-to-head
$375.00 against $187.50
A month of moderate agent traffic: 50M input tokens and 5M output tokens on a flagship Claude model.
OpenRouter passes through provider list price, which is $5.00/M input and $25.00/M output — 50 × $5 + 5 × $25 = $375. We charge half list, so the same traffic is $187.50. On top of that, topping up $375 of OpenRouter credit by card costs about $20.63 in their 5.5% purchase fee, so the real gap is closer to $208. Their fee schedule is published in their own FAQ; we are not estimating it.
Where OpenRouter wins
- +Widest model selection (200+)
- +Automatic fallback across providers
- +Free tier for many open models
- +Well-known brand
Where OpenRouter loses ground
- −Token rates are pass-through list — $5.00/M input on Opus 4.7, no discount
- −5.5% fee on credit purchases ($0.80 minimum), 5% on crypto, so effective cost sits above list
- −Model instance quality varies by provider pool unless you pin a provider
What you give up by switching
Stated plainly, because you will find this out in week two anyway and it is cheaper for both of us if you find out now.
- !We carry a focused model list — Claude, GPT, Gemini and Fable — not 200+. If your work depends on open-weight models, OpenRouter covers ground we do not.
- !There is no free tier here. OpenRouter serves a number of open models free, rate-limited to 50 requests a day without credit on the account and 1,000 with it.
- !No automatic cross-provider failover. We are a single upstream path per model, so a model being down means it is down, and our status is what our patch log says it is.
- !They are the better-known brand with a larger community. If you want a Discord full of people who have hit your exact problem, that is there and not here.
Moving over, step by step
- 01 Change the base URL
Point your client at https://api.llmrelay.dev/v1 instead of https://openrouter.ai/api/v1. Both are OpenAI-compatible, so the request and response shapes are unchanged.
- 02 Swap the key
Replace your OPENROUTER_API_KEY value with an llmrelay key. Most SDKs read a single environment variable, so this is one line in your config.
- 03 Drop the vendor prefix from model ids
OpenRouter namespaces models as anthropic/claude-opus-4.5. We use the plain upstream id, so that becomes claude-opus-5 for the current flagship. This is the only change that touches code rather than config.
- 04 Delete the HTTP-Referer and X-Title headers
Those are OpenRouter-specific attribution headers for their model rankings. We ignore them, so they are harmless if left in place, but they no longer do anything.
Questions people actually ask
Is llmrelay cheaper than OpenRouter?
On the premium Claude, GPT and Gemini tiers, yes, and by a wide margin. OpenRouter passes through provider list price without an inference markup, so a flagship Claude model costs $5.00 per million input tokens there. We charge 50% of list, so the same model is $2.50. OpenRouter also adds a 5.5% fee when you buy credit by card, which we do not. Where OpenRouter can be cheaper is the free tier it offers on many open-weight models — we have no free tier.
Does OpenRouter mark up inference?
No, and it is worth being precise about this because it is a common misreading. OpenRouter genuinely passes through the underlying provider price on inference. Their revenue on paid models comes from the credit-purchase fee: 5.5% with a $0.80 minimum on card payments, 5% on crypto, and 5% on bring-your-own-key usage above their monthly request allowance. So the comparison is not markup against no-markup, it is list price against half list.
What does OpenRouter do that llmrelay does not?
Breadth and routing. They front 200+ models including many we do not carry, and they will automatically fail over between providers hosting the same model. If you need an obscure open-weight model, or you want that cross-provider failover as a product feature, they are the better tool and we would tell you so.
Can I use both at once?
Yes, and for some teams that is the right answer. Both speak OpenAI-compatible, so you can route premium Claude and GPT traffic through us for the price, and keep OpenRouter for long-tail models. The only thing to watch is that model ids differ — they namespace with a vendor prefix, we do not.
Figures for OpenRouter come from OpenRouter's own documentation, checked 31 July 2026. If they have changed their pricing since and we have not caught it, the error is ours — check their page before you decide.
Migrating from OpenRouter?
Both providers speak OpenAI-compatible / Anthropic-compatible. Swap two env vars and your existing app keeps working.
Get API key →