One gateway. Every model. Your data stays yours.
Private inference on 200+ models, guarded on every call. Zero markup on pay-as-you-go and subscriptions. Or bring your own keys.
No credit card · live in 60 seconds · prompts aren’t logged unless you turn it on
from openai import OpenAIclient = OpenAI(base_url="https://api.orcarouter.ai/v1",api_key=ORCAROUTER_API_KEY,)resp = client.chat.completions.create(model="orcarouter/auto", # we grade + routemessages=[{"role": "user", "content": "..."}],)
One line. We grade each prompt, route to frontier or OSS, and add $0.
Private by architecture. Guarded at every call.
What you ask an AI stays yours. What your agents do is yours to approve.
Your data
What the model and our servers can see.
Data Cloaking
Cloaked.
Names, emails and card numbers are swapped for stand-ins before they reach the model you call, and put back in the answer.
One clickZDR
Nothing kept.
Prompts aren’t logged unless you turn logging on. Sign a zero-retention agreement to lock it off.
DefaultPQC
Quantum-safe.
Logs you choose to keep are sealed with hybrid ML-KEM. The crypto is open source: SCUTTLE.
DefaultData Residency
Your region.
Declare US, EU, UK, Asia-Pacific or China, and your compliance reports are tagged with that region.
On requestTEE
Sealed while it runs.
Attested enclaves, with proof for every answer.
On request
Your agents
What gets through, and what runs.
Guardrails
Blocked before it’s billed.
PII, secrets, jailbreaks and prompt injection stopped before the model sees them. Pattern rules are free; an AI-judge rule bills only its own check.
One clickAgent Firewall
Every tool call, graded.
Each tool and MCP call is allowed, sent to a person for review, or blocked before it runs.
Watching by defaultCompliance Packs
Mapped to 32 frameworks.
One policy per regulation. Watch it on real traffic first, then turn enforcement on.
Paid plans
Also:Human approval queueThreats tagged to OWASP LLM Top 10 and MITRE ATLASThe public archive of agent incidents
Route smarter. Spend less.
Every model. One price list.
The newest of 200+ models, live, at the provider's own price.
| Model | Routed to | Input /M | Output /M | Context | Private route |
|---|---|---|---|---|---|
| openai/gpt-6.1-solTextNEW | OpenAI Direct | $2.00 | $10.00 | 1M | No private route |
| anthropic/claude-sonnet-5.5TextNEW | Anthropic Direct | $2.00 | $10.00 | 1M | No private route |
| typesafe/jev-1.13TextNEW | — | $0.042 | $0.042 | 66K | No private route |
| openai/gpt-6-lunaTextNEW | OpenAI Direct | $0.100 | $0.500 | 1M | No private route |
| openai/gpt-6-solTextNEW | OpenAI Direct | $2.00 | $10.00 | 1M | No private route |
| anthropic/claude-opus-5.5TextNEW | Anthropic Direct | $4.00 | $20.00 | 1M | No private route |
| grok/grok-4.7TextNEW | — | $2.00 | $6.00 | 500K | No private route |
| orca/orcaverify-text1.0TextNEW | — | $2.00 | $2.00 | — | No private route |
| + 194 more models · prices update every 60 seconds | |||||
Live in 60 seconds.
Point your SDK at us
Set base_url to api.orcarouter.ai/v1 and swap your API key. No other code changes needed.
We route, guard and observe
Graded in under 1ms, with failover, caching and full logs built in.
You ship, on one endpoint
Direct to each provider at their published rate — we add $0 per token.
We never mark up the tokens you buy.
Our revenue comes from optional team features.
Hacker
Team
Enterprise
Three ways to pay for tokens.
Top-ups and subscriptions bill at provider price. We add $0.
Pay as you go
Top up when you want. Provider price, receipt on every call.
Credit subscription
Refills your wallet each period. Full amount, any model, plus bonus credit on larger plans.
Bring your own key
Use your own provider keys, rate limits and credits.
Plans above are for team features. They don't change token price.
Fresh from the engine room.
What we're building and why — our latest posts.
Privacy, security and routing, answered.
Do you store or train on my prompts?
We never train on your prompts. Prompt and response bodies aren't logged unless logging is on for your workspace; we keep request metadata (time, model, tokens, cost, IP) for billing and security. A zero-retention agreement locks body logging off.
Where is my data processed?
By the provider that serves the model you pick, in that provider's regions. Your workspace's data-region setting (US, EU, UK, Asia-Pacific or China) tags your compliance reports with that region; it doesn't yet pin where your data is stored or where inference runs.
What does quantum-safe mean here?
Request logs you choose to keep can be sealed with hybrid post-quantum encryption (ML-KEM-768 + X25519), so a copy taken today can't be opened by a future quantum computer. The code is open source: SCUTTLE. Sealing is on by default on our hosted service.
What is an AI router?
An AI router sits between your app and many models and picks the best model for each request. OrcaRouter grades every prompt in under 1 ms and routes it — frontier for hard reasoning, open-source for routine — across 200+ models at zero token markup.
What is an LLM router and how does it work?
An LLM router classifies each prompt and matches it to the model most likely to answer well at the lowest cost. OrcaRouter routes on contextual embeddings with online learning from live traffic — 75.5% accuracy on the public RouterArena leaderboard.
Is OrcaRouter an AI gateway?
Yes. OrcaRouter is a production AI gateway: one OpenAI-compatible endpoint with adaptive routing, load balancing, automatic failover, guardrails, an agent firewall, prompt caching and per-request observability.
What is an AI CDN?
Like a CDN serves content from the best edge, an AI CDN serves inference from the best provider: healthy, fast, cheap capacity, cached repeated prompts, failover on outages. OrcaRouter plays this role across 200+ models.
What is adaptive routing?
Adaptive routing learns the quality/cost trade-off from your own traffic. Point each workspace at cheapest-that-clears-the-bar, highest quality, or balanced — or let orcarouter/auto keep tuning the choice per request.
Do I need both an AI gateway and an LLM router?
They solve different problems — governance vs model choice — but you don't need two systems: OrcaRouter routes per request and enforces budgets, guardrails and observability on the same hop.
Does routing add latency?
Prompt grading takes under 1 ms and total added latency stays under 50 ms — usually won back many times over by faster providers and cached prompt tokens.
Does OrcaRouter mark up token prices?
No. You pay each provider's published rate; OrcaRouter adds $0 per token and monetizes optional Team and Enterprise features.
Can I keep my existing OpenAI SDK code?
Yes — switch base_url to https://api.orcarouter.ai/v1 and your OpenAI, Anthropic or Google SDK code keeps working: Chat Completions, Responses, Embeddings, Images, Audio and streaming.
What happens when a provider goes down?
OrcaRouter retries against healthy fallback capacity for the same or an equivalent model before the response starts, so upstream outages don't surface to your users.
Private. Smarter. Zero markup.
Swap one line. That's the migration.
Beats GPT-5 & Azure on RouterArenaBacked by published researchRead the Scuttle source (opens in new tab)Visit the Trust Center
OrcaRouterFor Providers
Run an inference platform? Get your models on OrcaRouter.
providers@orcarouter.ai