The whole migration
Two lines. That's it.
Point the base URL at Parity Layer and use your Parity key. The API is OpenAI-compatible, so your SDK, streaming, and function calling keep working. Your traffic keeps running on your current model, untouched, while the proof builds in the background.
# your existing client
- base_url = "https://api.openai.com/v1"
+ base_url = "https://api.paritylayer.com/v1"
+ api_key = os.environ["PARITY_API_KEY"]
# everything else stays the same
The business model
Why is the router free?
Because the router is not the product. The proof is.
The router: free
Connect your traffic to our OpenAI-compatible API, or start with just a JSONL export. Routing costs nothing. You connect your current provider once and we handle every provider we route to after that.
The product: the proof
While you route, our patent-pending engine tests cheaper models against your real prompts. A blind judge, your own model class judging, decides when one produces results as good or better than your current model.
The price: one simple number
When a cheaper model is proven, you see the proof and a single integrated per-token price for running on it, typically 30 to 60% below your current bill. Switch in one click, instant fallback if quality drifts. No proof, no charge.
What this is not.
- A 300 model menu
- Multi-provider failover
- Load balancing
- Coding agents
We route to your baseline, and after a proof, to the one cheaper model that earned it.
Straight answers
Free router, straight answers
Is it really free to route through Parity Layer?
Yes. Routing through Parity Layer costs nothing, and you do not bring any API keys: connect to our OpenAI-compatible API, or start with a JSONL export of past requests, and we handle the provider side end to end. If our proof later shows a cheaper model producing better results on your prompts, you get the proof and one simple integrated price for switching, billed per request and per token. If nothing proves out, you pay nothing, forever.
Is this a router with a big model menu?
No, and we say that plainly. There is no 300 model menu, no multi-provider failover, no load balancing. Parity Layer routes your traffic to your baseline model, and after a proof, to the one cheaper model that was proven to match or beat it on your own prompts.
What does the quality proof actually do?
In the background, cheaper candidate models run against your real prompts. A blind judge, using your own model class as the judge, decides whether the cheaper output matches or beats your baseline. Only a proven model is ever offered as a switch, and fallback is instant if quality drifts.
What do I change in my code?
Two lines: point base_url at api.paritylayer.com and use your Parity key. The API is OpenAI-compatible and works with OpenAI, Anthropic, and Google SDK shapes.
Does it work for coding agents?
No. Coding agents are not our workload.
Route free today. Get proof this week.
Two lines of config and the proof engine starts working on your real prompts. Or upload a JSONL export of past requests and see the proof without routing anything.