Availability SLA
99.9%
Gateway requests · monthly measurement window
Monthly availability target for gateway requests.
PRODUCTION LLM GATEWAY
A production-readyLLM API gateway.
Access GPT, Claude, Gemini and 50+ models through one OpenAI-compatible API, with configurable provider fallback, usage controls and request-level visibility.
- OpenAI-compatible
- Configurable fallback
- Pay as you go
- Save up to 25%
One API for leading model providers
Three-line integration
Keep your OpenAI client. Change the base URL.
RouteAPI uses an OpenAI-compatible interface, so most integrations can keep the existing SDK and update the API key and base URL.
Supported API features may vary by model and upstream provider.
Read the API docsmain.py
BeforeOriginal
1from openai import OpenAI
2
3client = OpenAI(
4 api_key="sk-openai-key",# set ROUTEAPI_KEY in your environment
5)
6
7response = client.chat.completions.create(
8 model="gpt-4o",
9 messages=[{"role": "user", "content": "Hello"}],
10)
AfterChanged
1from openai import OpenAI
2import os
3
4client = OpenAI(
5 api_key=os.environ["ROUTEAPI_KEY"],# set ROUTEAPI_KEY in your environment
6 base_url="https://www.routeapi.ai/v1",# point the client to RouteAPI
7)
8
9response = client.chat.completions.create(
10 model="gpt-5.5",# use a model enabled for your account
11 messages=[{"role": "user", "content": "Hello"}],
12)
MULTI-MODEL ACCESS
One API for the models your application uses.
Send requests through one OpenAI-compatible endpoint while selecting the supported model your application needs.
RouteAPI
OpenAI
OpenAIAnthropic
AnthropicGoogle
GoogleDeepSeek
DeepSeekQwen
QwenMiniMax
MiniMaxMoonshot
MoonshotZhiPu
ZhiPuOther supported models
Other supported modelsMODEL ROUTING & RESILIENCE
One model route. Multiple provider paths.
Choose a model and configure its primary and fallback providers. RouteAPI monitors channel health and routes requests through the available path according to your policy.
Primary channel healthy
Your app
RouteAPI
claude-sonnet5Selected model: claude-sonnet5
Simulate a provider failure to see the next request move to the configured fallback.
Your application selects a model and sends requests through RouteAPI. You configure the provider order; RouteAPI monitors channel health and uses the configured primary or fallback path according to your policy. A provider failure affects the next request, not an in-progress streaming response.
Transparent pricing
Live model prices and availability from the catalog your workload uses.
25% below official API list prices
On our most-used US-based models
20% below official API list prices
On our most-used China-based models
No additional platform fee on top of the displayed model rate.
Billing principles: Pay only for billable usage from a completed request. A failed request with no billable usage is not charged; when a configured fallback completes the request, that usage is billed at the selected model's displayed RouteAPI rate.
Control plane
Control AI traffic from one operating layer.
Replace separate provider endpoints, credentials and operational records with one OpenAI-compatible control layer.
Platform engineeringAI application teamsOperations and finance
www.routeapi.ai/console
DashboardAll routes healthy
Incoming requests
0
+12.4%Today's tokens
0
TodaySuccess rate
0.00%
Cache hit rate 86.4%Estimated cost
$0.00
Saved $36.79Request volumeLast 24 hours
18,420 requests
29,680 requests
22,450 requests
38,120 requests
26,740 requests
45,200 requests
31,980 requests
48,600 requests
29,150 requests
40,780 requests
43,820 requests
51,240 requests
Direct provider integrations
- Every provider brings a separate endpoint, credential and billing surface.
- Provider outages and model changes reach your application directly.
- Usage and costs are split across multiple consoles.
RouteAPI gateway
- One OpenAI-compatible endpoint for your application.
- Route selection, retries and fallback behavior are managed in one place.
- Usage, latency, failures and spend are visible in one operating view.
RELIABILITY BASELINE
Clear service targets, measured by scope.
Public reference targets for the gateway layer. Model inference and streaming generation are measured separately because their latency varies by workload.
Gateway overhead (P99)
≤ 400 ms
RouteAPI processing only · model inference excluded
99% of eligible requests should add no more than 400 ms at the gateway layer.
Failover objective
≤ 1 s
Eligible provider or transport failures · next request
Internal recovery target; the objective is policy-defined because no universal failover time applies to every workload.
Provider integrations
30+
Supported upstream families · current product scope
Product coverage count, shown separately from availability and latency SLOs.
These figures are our service level targets and current product scope, not live production telemetry. Contractual terms are agreed per enterprise plan.
DATA HANDLING TRANSPARENCY
Understand the path your API data takes.
A concise view of where requests go and what RouteAPI records. Upstream providers apply their own data policies.
Your application
Sends prompts and request parameters
RouteAPI gateway
Routes the request and records operational data
Configured upstream provider
Processes the request under its own data policy
Where requests go
Requests pass through RouteAPI and are sent to the upstream provider configured by the system for the selected model.
Model training
RouteAPI does not store your request content and will not use it for training. Your data is forwarded directly to the selected upstream provider; please refer to each provider's data policy.
What RouteAPI records
Operational records include model, provider, status, latency, token usage and cost.
This page describes RouteAPI's current request path and logging behavior; it does not make a blanket guarantee for upstream retention, deletion or training policies.
Questions, answered
What teams need to know before routing production traffic.
Five practical answers for your first production call. Deeper details stay in the docs and pricing catalog.
Do we need to replace our existing OpenAI client?
Usually not. Keep the OpenAI client, update the Base URL and API key, and check the compatibility matrix because supported features can vary by model and upstream provider.
Does RouteAPI automatically choose the best provider?
Yes. You select the model you want to use, and RouteAPI automatically chooses the most suitable provider based on the system-configured route and provider status.
How do provider fallback and retries work?
You configure a primary provider, fallback providers and retry policy for a model route. RouteAPI applies those rules when a qualifying provider or transport failure occurs; not every error can be recovered.
Are failed requests, retries and fallback attempts charged?
Billing follows billable usage. A failed attempt with no billable usage is not charged; a retry or fallback that generates billable usage can be charged, and a fallback that completes the request is billed at the selected model rate. Read billing details
Do you store prompts and responses?
No. RouteAPI does not store prompt or response content. For billing, we retain only necessary metadata such as the model, token usage, cost, request status and timestamp.
Send your first request through RouteAPI.
Create an API key, update your base URL and send a test request.
Pay as you go · Save up to 25%
Trusted at scale
The teams building on RouteAPI
A production-ready gateway for teams that need reliable access to the models behind their products.