Platform
MCP Server Build
A production MCP server for one of your systems, with up to 8 tools, secure sign-in, action logging and tests in AI…
Platform
AI Model Router
Each request sent to the model best suited to it, weighed by cost, capability and latency, automatically.
Sending every AI request to your most powerful model is expensive, and sending everything to the cheapest one hurts quality. AI Model Router weighs each request against rules you set on cost, capability and latency, then sends it to the model that fits. Fallbacks cover provider outages. Every decision and its reason is logged, so you can see why each request went where it did.
- Simple requests stop running on your most expensive model
- Fallback rules keep requests flowing when a provider struggles
- Every routing decision logged with the reason behind it
$3,500
20–30 working days
Optional maintenance $249/mo
What it is
AI Model Router is a build that decides which AI model handles each incoming request, weighing cost, capability and latency against rules you set, so simple tasks don't run on your most expensive model and complex ones don't run on your cheapest.
Why businesses buy it
Most AI applications pick one model and use it for everything. That means paying premium rates for simple tasks, or getting weak answers on hard ones, and no easy way to change course. After this build, each request is matched to the right model by rules your team controls, with the decision recorded for review. You can tune cost and quality without rewriting applications. The Scope Card sets the applications, routing rules and controls in writing before you buy, so you know exactly what will be configured.
Problems this solves
Simple tasks run on your most expensive AI model
The Model Router decides which model handles each request, weighing cost, capability and latency against rules you set. Routine requests go to a cheaper model.
Complex requests get a weak answer from a cheap model
Routing rules send harder tasks to a more capable model. Up to five rules or controls are configured from your policies, across up to three applications.
Changing which model handles a task means editing code
Routing decisions live in the router's configuration, not in each application. The admin guide shows your team how to adjust rules as models and prices change.
A good fit if
You call more than one model across your workflows and currently pick which one manually or hardcode it per application.
Some of your requests are simple and overpaying for a top-tier model on them is wasting budget every day.
You have rules in mind, such as cost ceilings, capability requirements or latency targets, that should decide routing automatically.
You want routing decisions logged, so you can see which model handled which request and why, per workflow.
Not the right fit if
You need the full integration layer, keys, rate limits and logging for every provider, not just routing logic. See AI Gateway.
Your priority is versioning and testing prompts, not deciding which model runs them. See AI Prompt Management.
You need agents to retain memory across sessions, not routing decisions per request. See AI Agent Memory System.
You need structured retrieval from your own documents, not model selection. See AI Agent Knowledge Layer.
How it works
- We map your current workflows and the models each one calls, along with any manual routing decisions already being made.
- Routing rules are defined against your priorities: cost ceilings, capability requirements such as reasoning or context length, and latency targets.
- The router evaluates each incoming request against those rules and sends it to the model that fits, across providers such as OpenAI, Gemini or Bedrock.
- Fallback logic is configured, so a request routes to a backup model if the preferred one is unavailable or over quota.
- Every routing decision is logged with the model chosen and the reason, exportable to Langfuse or Grafana for review.
- A control test log documents routing behavior against real requests, reviewed with your technical owner before launch.
What you provide within 72 hours
Your portal lists these the moment you order. Delivery starts the working day after everything is in and your access is checked.
- Written authorization confirming which applications and provider accounts are in scope for routing decisions.
- A list of the AI applications and systems whose requests should be routed by the model router.
- Your routing priorities, such as cost ceilings, required capabilities, or latency targets per workflow.
- Cloud or hosting access needed to deploy the router alongside your existing infrastructure.
- A named technical owner who can review routing rules and sign off before launch.
- Admin access and API keys for every connected provider, shared by adding our team as a user or creating a key you control, never by email.
- Your business information (company basics, hours and brand voice), shared during intake, and one decision-maker who can approve the final configuration.
The 72-hour rule. Send everything on your checklist, including access, within 72 hours of ordering so your build starts on time.
Timeline
Start
Intake verified, workflows and routing priorities confirmed.
Weeks 1–2
Routing rules built and fallback logic configured across your provider accounts.
Week 3
Applications connected to the router and tested against real requests.
Weeks 3–4
Control test log produced and shared with your technical owner.
Week 4
Your review and two rounds of changes applied to the routing rules.
Go-live
Live, recorded handoff session, 30 days of support begins.
| Option | Delivery time | Price |
|---|---|---|
| Standard | 20–30 working days | $3,500 |
| Express | 15–25 working days | $4,100 |
| Rush | 10–20 working days | $4,750 |
What you receive at handover
- AI Model Router configured and deployed, covering up to three systems or AI applications
- Routing rules weighing cost, capability and latency (up to five controls)
- Routing reporting in an admin console or written report
- A test report with a control test log for the agreed scenarios
- An admin guide and a recorded 60-minute handoff session
- 30 days of post-launch support
Expected benefits and how to measure them
- Can lower AI spend by matching each task to a suitable model.
- Helps keep answer quality steady on harder requests.
- Makes model changes a configuration task rather than a code release.
- Cost per request
- Divide monthly provider spend by request volume, before and after launch.
- Requests by model
- Read the share of traffic per model from the router report each month.
- Quality on routed requests
- Review a sample of answers from each model monthly against your own quality checklist.
Before you start: Record last month's provider spend and request volume per application. Note which model each application uses today.
These are things to track, not promised results. Outcomes depend on your data, your team and your market.
Running costs
The build price is one-time. Running the Model Router afterward costs roughly $20 to $300 a month in your own accounts, covering hosting, logging and monitoring tools, and it typically lowers your overall model spend by avoiding overpowered models on simple requests.
AI model usage during the build and testing is billed to your own accounts too, so every provider relationship stays yours, not ours.
Ask about running costsCustom requirements
Need more than three systems in scope, more than five routing rules configured, or a second admin console? Add a note at checkout and you get a written Change Quote, only starting once you accept it. Larger or bespoke projects go through the Custom AI request form instead.
Request a custom versionKeep it running
Optional monthly maintenance covers monitoring, fixes, model and platform updates, a monthly health report and monthly change hours, enough to tune a routing rule or add a new model to the pool. It starts once the 30 days of support end and can be cancelled any month.
$249/mo (the minimum fee, includes 2 change hours)
Optional. Starts after the post-launch support period. Cancel any month.
How maintenance worksHow this compares
| Do it yourself with a SaaS tool | Hire a custom agency | Upmark Labs | |
|---|---|---|---|
| Price | A model routing platform charges a monthly fee per request volume | A systems integrator quotes per project after a discovery phase | Ours is one fixed price, shown on this page before you buy |
| Scope | A routing platform is infrastructure your engineers configure and tune | An integrator agrees scope during the project | Ours is written on the Scope Card before purchase |
| Rule design | A routing platform ships default rules you adjust yourself | An integrator build depends on who is staffed on it | Ours is built against the cost, capability and latency priorities you set |
| Delivery | A routing platform is live once your team finishes configuring it | An integrator usually takes several weeks to staff and start | Ours has a fixed delivery time shown before you buy, with faster options |
Questions about this product
How much does AI Model Router cost?
The fixed price for each delivery speed is shown in the price and delivery panel on this page before you order. It covers everything on the Scope Card: routing rules for up to three applications, fallback logic, up to five controls or monitors, and the control test log you review against real requests before launch.
How does AI Model Router decide which model to use?
It evaluates each incoming request against the rules you set: cost ceilings, required capabilities such as reasoning depth or context length, and latency targets. Requests that fit a cheaper, faster model route there, while requests that genuinely need more capability route to a stronger model, with the decision and reason logged for review.
What does AI Model Router cost to run each month?
Roughly $20 to $300 a month in your own accounts, covering hosting, logging and monitoring tools such as Langfuse or Grafana. You pay providers directly, so there is no markup from us. Model usage itself bills to your provider accounts, and the routing logs show you which models that spend went to.
How is AI Model Router different from AI Gateway?
AI Model Router is the decision logic for which model handles a request, by cost, capability and latency. AI Gateway is the broader integration layer: keys, rate limits and logging for every provider your applications call. The two fit together, and the Router's logic can run as one component inside a Gateway deployment.
Can we change the routing rules after launch?
Yes. Adjusting an existing routing rule, such as a cost ceiling or a latency target, fits inside the monthly change hours included with maintenance. Adding a new provider to the routing pool or a larger rule set change is a bigger job and gets a written Change Quote first.
Can we add more systems to AI Model Router later?
Yes. The standard build covers up to three systems or AI applications, and adding a fourth or beyond is quoted as a Change Quote once you are ready to expand. We suggest routing your highest-volume workflows first, confirming the cost and latency impact, then extending routing to further applications.
What do we need to provide before AI Model Router can launch?
Written authorization for the applications and provider accounts in scope, a list of AI applications whose requests should be routed, and your routing priorities on cost, capability and latency. We also need cloud access, a named technical owner, and admin access shared by adding our team as a user or creating a key you control, never sent by email.
Can I cancel maintenance for AI Model Router?
Yes, any month. Maintenance starts after the 30 days of post-launch support end and runs month to month with no cancellation fee and no lock-in. It includes a monthly allowance of change hours that scales with the build, and bigger changes go through a Change Quote.
Stop paying premium model prices for requests that don't need them.
Order AI Model Router and every request routes to the right model on cost, capability and latency on a fixed delivery date. The scope above is what you get, in writing, before you pay.
Works well with
Platform
AI Agent Orchestration System
Coordinates multiple agents, tools and handoffs in one runtime.
