llmrelay vs OpenRouter
OpenRouter is an excellent model marketplace that sells tokens at provider list price. llmrelay is a focused Claude/GPT/Gemini relay at half that. Different jobs.
Where the money goes
On Claude Opus 4.7, per million input tokens:
Credit for the accurate part: OpenRouter does not mark up inference. Its own FAQ says it passes provider pricing through without markup. The gap is not a hidden fee on their side, it's a discount on ours — we buy capacity in bulk and resell every model at half of official list.
One fee does apply on top: buying credits costs 5.5% (Stripe, $0.80 minimum) or 5% via crypto. So $100 of usable OpenRouter credit costs about $105.50, putting effective Opus 4.7 input near $5.28/M. Bring your own provider key and the first tranche of BYOK requests each month is free, then it's 5% of what the call would have cost.
Model coverage — where OpenRouter wins
OpenRouter routes 200+ models across dozens of providers. If you want obscure open models — Kimi K2, DeepSeek variants, older Llamas, community fine-tunes — OpenRouter is where they live.
llmrelay is deliberately focused: Claude family, GPT family, Gemini family, Fable family. If you don't need Kimi at 3 AM, you're paying list price for a marketplace you aren't using.
Billing model differences
- — Token rates are pass-through list, no inference markup
- — 5.5% fee on credit purchases ($0.80 min), 5% on crypto
- — BYOK costs 5% of the equivalent call after the free monthly allowance
- — Model instance quality varies by provider pool unless you pin one
- — One prepaid balance in USD, no expiry
- — 50% off official list on every model
- — Single upstream tier (no pool roulette)
- — Hidden-prompt neutralizer patch published in fork
Migration cost: zero
Both providers speak OpenAI-compatible and Anthropic-compatible APIs. Migration is two env var changes.
# Before (OpenRouter) OPENAI_BASE_URL=https://openrouter.ai/api/v1 OPENAI_API_KEY=sk-or-... # After (llmrelay) OPENAI_BASE_URL=https://api.llmrelay.dev/v1 OPENAI_API_KEY=sk-llmrelay-...
Cut your Claude / GPT / Gemini bill in half
If your load is 80% Claude/GPT/Gemini and 20% niche models, keep OpenRouter for the niche and put the rest on llmrelay. Or just move everything.
Get API key →