Neurometric Task Router
One API.
Intelligent Routing
To Any Model
Task Router starts off routing you to frontier models, then analyzes your traffic and recommends models that are faster, cheaper, and more accurate based on your unique workloads. If you want to use the best models, use Task Router.
Change your base_url, keep your code. No evals to build first.
Routing across models from
- Scaledown
- Neurometric
- Microsoft
- Alibaba
- Mistral
- IBM
How It Works
- 01Swap in one endpointReplace your existing model calls with the Task Router API endpoint.
- 02Nothing changes on day oneWe route to the models you were already using, at the same price and performance.
- 03We learn your workloadsWe analyze your traffic and recommend ways to improve latency, cost, and accuracy on your unique tasks.
- 04You stay in controlAccept the new routes, or keep what you have.
For developers
Two lines of change. Zero evals to build.
- 1
Point your OpenAI SDK at Task RouterKeep your client library and request code as they are.
- 2
Create a route for your taskDescribe what you need and we find or build the task for it. Set a minimum quality, and a latency cap if you need one.
- 3
Ship, then watch the route improveWith auto-optimize on, the route moves when a cheaper model qualifies.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.taskrouter.com/v1",
api_key=os.environ["TASKROUTER_API_KEY"],
)
reply = client.chat.completions.create(
model="taskrouter/privilege-triage",
messages=[{"role": "user", "content": email_thread}],
)Replace taskrouter/privilege-triage with your route's alias.
Not ready to route yet? Exploring tasks is free, with no account or API key. Describe a task and compare measured quality, cost, and speed across models before you commit.
Questions developers ask
What if my task isn't supported?
Describe it and we'll map it to the closest measured task. If nothing fits well, we'll benchmark it.
Is it really OpenAI-compatible?
Yes. Use the OpenAI SDK you already have, change the base URL, and set the model to your route's alias. Your client code stays the same.
What happens if no model hits my target?
The route isn't created, so you never end up on a model that quietly misses your quality bar. Lower the target, relax the latency cap, or request a benchmark.
How is this different from OpenRouter?
Unified API routers give you access to many models. Task Router also decides which model to use, based on measured quality for your specific task.
Can I control how a route picks models?
Yes. Each route has a preset (lowest cost, balanced, or highest quality), a minimum quality and a p50 latency cap. Turn off auto-optimize to hold a route steady.
How fresh is the pricing data?
Provider prices are observed daily across routers. Every price on a task page shows the date it was observed.