Route models on your terms.
One OpenAI-compatible endpoint. Route to any major model by cost, by latency, or by a rule you define. Model pricing passed through at cost, platform fee listed separately.
Intelligent routing for every workload
Not every request needs the same routing intelligence. A production API call should route through full evaluation — cost, latency, availability. A CI test needs zero overhead. An overnight batch job just needs the cheapest working model. Three layers exist because the routing decision should match the request — not the other way around.
Three layers for every kind of intent
L1 — Auto-routing
Send a request. We evaluate cost per token, P50 response time, and provider uptime, then route to the best fit. No manual tuning required.
"I just want it to work."
L2 — Manual
Specify the model in your request — model: deepseek-chat. We forward it exactly. No substitution, no smart routing, no surprises.
"I know exactly what I want."
L3 — Custom Rules
Define routing logic as conditions. Task routing, time-based routing, budget-aware routing — compiled and executed per request. No middleware to build.
"My infrastructure should follow my logic."
Enable one layer or all three per request. The integration code stays the same.
Routing comes with overhead.
Ours earns it back.
Provider health checks, latency scoring, rule evaluation — each request adds a few milliseconds. Combined with caching and smart routing, the routing layer typically reduces total API spend by 20-40%.
| Direct to provider | With GMH routing | |
|---|---|---|
| Model cost | $X per token | $X per token (pass-through) |
| Platform fee | — | Flat % on usage |
| Total spend impact | — | −20–40% of bill |
Model cost passed through at cost. Flat platform fee listed separately. No markup on inference. Here's how the numbers work out in practice.
Cost breakdown: over 12 months
Two scenarios illustrate the difference. One pays per-request without routing intelligence. The other uses smart routing and caching — the spend curve flattens over time.
Direct API vs GMH
Over 12 months, the gap widens. Direct API costs scale with usage; GMH's routing and caching keep spend flat.
Savings accumulation
Caching and intelligent routing compound each month. The routing layer pays for itself by month 3.
Common questions
A few things people ask about routing, pricing, and how BYOK works.
Key features, summed up.
Pass-through pricing
Model cost passed through at cost — no markup on inference. Platform fee listed separately. You see exactly what you're paying for routing versus inference.
20-40% savings
Smart routing sends simple queries to cheap models and complex ones to premium — only when needed. Combined with caching, the routing layer typically reduces total spend by 20-40%.
One-line integration
Your existing OpenAI SDK works as-is. Change base URL and API key — streaming, tool calls, vision, structured outputs all work. No new dependencies.
Every major model
Switch models by changing a string, not your integration.
Start routing in one line.
OpenAI-compatible. Free tier. No credit card required.