Overkill by default
Eight in ten support questions are as simple as "How do I return this?" Yet every single one is billed at flagship-model rates.
Smart Routing
The AI gateway that cuts your bill 20–40%
MixRoute Smart Routing reads every request in real time with our in‑house AI model, judging complexity and task type, then sends it to the best-fit model in the pool you authorize.
- 10,000+ real-world runs
- 10,000+ real-world runs
- Avg. savings 20–40%
- Avg. savings 20–40%
- Free for a limited time
- Free for a limited time
- Zero data retention
- Zero data retention
How do I return this?
Summarize this meeting
When do you close?
Refactor this module
Smart Routing
real-time triage
SIMPLE
Gemini
COMPLEX
Claude
ULTRA
GPT
Gemini
COMPLEX
Claude
ULTRA
GPT
MEASURED RESULTS
Results from 10,000+ real-world runs
We tested Smart Routing's assignment decisions across tens of thousands of real tasks: the average saving came out at 20–40% of API cost. The exact figure depends on the vendor mix in your model pool, which is why we quote a measured range instead of a marketing number.
See what you'd save
Monthly API spend
US$5,000
US$500
US$50,000
Estimated monthly savings
—
Estimated from the average range measured across 10,000+ runs. Actual savings vary with your model pool.
You're paying flagship prices for questions that don't need one
Set once, never touched
Engineers pick one model per feature at launch, and nobody ever revisits the choice. Tasks keep changing. The model never does.
Updates you can't chase
The model market shifts every month. Last month's best price-performance pick may already be beaten. No human keeps up.
The problem isn't the models. It's how they're assigned.
HOW IT WORKS
Three steps, fully automatic
1
Step 1
Call as usual
Zero code changes. Create a Smart Routing API key in the console, swap out your current key, and keep sending requests in the same format.
client = OpenAI(api_key="sk-your-current-key",api_key="sk-smart-routing-key",base_url="https://api.mixroute.ai/v1")# request format unchanged
2
Step 2
Real-time triage
Our in-house routing model reads each request on the fly and judges its complexity and task type. It supports a 1M-token context window, so even the longest requests are fully assessed. Read-only, never stored: content is processed in memory and cleared the moment triage ends, fully in line with MixRoute's zero data retention policy.
- What it reads
- Complexity × task type
- Context window
- 1M tokens
- Where it runs
- In-memory only
- After triage
- Cleared instantly
- Data retained
- Zero
3
Step 3
Best-fit assignment
From the pool you authorize, the router picks the model best suited to the request. Your full context passes through untouched: no rewriting, no compression, no truncation. The answer still comes from the real model.
- gemini-3.5-flash standby
- claude-sonnet-5 best fit
- deepseek-v3.2 standby
- gpt-5.6-sol standby
Full context passes through untouched. Answers come from the real model.
CONTROL
AI runs the routing. You set the boundaries.
Fully automatic, never a black box. Smart Routing only ever operates inside boundaries you define: the model pool you authorize, the limits you set, the savings you can see.
Your model pool, your call
Give every key its own model pool in the console. Which model endpoints can be assigned is entirely up to you.
Limits you set
Set usage limits on every key; anything beyond them is blocked. No runaway requests, no runaway bills.
Savings you can see
A built-in savings dashboard shows exactly how much Smart Routing has saved you. Not our claim. Your numbers.
Simple
Light models
Google
High-volume light requests: classification, extraction, FAQ.
Complex
Core models
Anthropic
Multi-step reasoning, code, long-form analysis.
Ultra
Flagship models
OpenAI
The deepest reasoning and critical decision tasks.
Smart Routing has saved you
US$1,284
31.4% saved
Spent
Saved
Illustrative screens
USE CASES
Where teams feel it most
Scenario 01
High-volume support bots
Masses of repetitive, simple questions are Smart Routing's biggest source of savings. Easy ones go to light models, hard ones auto-upgrade. Customers never notice. Your bill does.
Biggest savingsScenario 02
Internal knowledge assistants
Policy lookups, document search, meeting summaries: most requests don't need a flagship model, but the occasional deep dive does. The router tells them apart, so you don't have to.
Depth, auto-detectedScenario 03
AI products with mixed workloads
You can't predict what users will ask, and simple and complex arrive interleaved. Instead of betting everything on one model, give each request exactly the model it needs.
Right-sized requestsScenario 04
Coding assistants
Completions and small functions run fast and cheap on light models; cross-file refactors and gnarly debugging auto-upgrade to a flagship. Engineers write code. The router picks models.
Auto gear-shiftPRICING
Try first. Decide later.
US$0
Free for a limited time
Activate now and Smart Routing costs nothing during the launch window. Validate the savings on your real traffic; the dashboard numbers will give you the answer.
5%
Then 5% per routed request
After the free period, we charge a 5% service fee only on requests that actually go through Smart Routing. Requests on regular keys are never charged.
THE ONE-LINE MATH
Pay 5%. Save an average of 20–40%.
You save far more than you pay. That's the entire reason this exists.
FAQ
Frequently asked questions
Does Smart Routing add latency?
Yes. Triage adds one extra step, and the time varies with request length. It’s a trade: a little time for 20–40% off your costs. For latency-critical endpoints, use a regular key to call a specific model directly and skip the router.
Does the routing model see my data?
It reads, it never stores. Request content is classified in memory in real time, cleared immediately afterward, and passed through to the target model untouched. Nothing is written to persistent storage, consistent with MixRoute’s zero data retention policy.
What if the router gets it wrong? Can I force a specific model?
Your model pool is the insurance: the router only ever chooses among models you’ve authorized and never assigns anything outside the pool. If some traffic must always hit one specific model, create a separate regular key for it and call that model directly.
How is this different from auto-routing on other platforms?
One word: control. Most auto-routers are a total black box where the platform decides everything. MixRoute’s Smart Routing runs inside a model pool you define, with zero data retention and a savings dashboard. You always know where the money is saved and where your data goes.
Do I need to change my code?
No. Create a Smart Routing API key in the console and swap it in for your current key. The API call format stays exactly the same.
How does pricing work?
Free for a limited time. Once the free period ends, only requests that go through Smart Routing are charged a 5% service fee on that request’s cost. It’s deducted from your credit balance, with a transparent, itemized bill.