We have a big announcement today - Taskrouter.com is officially out of beta.
AI has moved from the playground to production, and the bills moved with it. Teams now spend thousands of dollars sending every query to a massive frontier model, including the queries that are simple.
Most developers know this is wasteful. You want frontier-level reasoning for the complex edge cases. For the repetitive tasks that make up most of your traffic, you want the speed and cost of a specialized small language model (SLM).
The hard part is knowing which is which. Deciding when a smaller model is good enough takes constant benchmarking, testing and custom routing logic. That is a full-time job, and it isn’t the product you set out to build.
Taskrouter does that job for you. It isn’t another API aggregator. It’s a router that learns your specific tasks and optimizes them for quality, speed and cost.
How Taskrouter works: from frontier to specialized #
Integration is one API key. You point your application at Taskrouter the same way you would at OpenRouter or OpenAI.
From there, four things happen:
- Start strong. Your initial workloads go to top-tier frontier models. That sets a quality baseline and captures gold-standard outputs for each task.
- Test continuously. Behind the scenes, Taskrouter runs those same tasks against a fleet of smaller, specialized and open-source models.
- Hand off when the data says so. Once a cheaper, faster SLM is shown to match or beat the frontier model on your specific task, Taskrouter routes that traffic to it automatically.
- Keep the safety net. Automatic failover to frontier models stays in place for tasks that require deep reasoning.
The result is lower cost and lower latency at the same quality. Our first Fortune 500 customer to deploy Taskrouter in production saved millions of dollars, over 90% of their inference costs, while seeing their latency drop by 200ms on average and accuracy improving 5 percentage points.
The elephant in the room: privacy, IP and your data #
When enterprise developers hear “learns from your data,” alarm bells ring. You picture your proprietary data being fed into a massive public model. That is a reasonable fear, and we built Taskrouter around it.
We come from enterprise. Taskrouter was built by the team that built my first company. We ran backup for Fortune 500 firms and had to pass rigorous security audits. In our current form we have been building SLMs for enterprise customers for over a year, and have active Fortune 500 customers on those SLMs. We know what it takes to pass a corporate infosec review.
Here is the Taskrouter privacy guarantee:
- It’s your IP. We don’t train public models on your data. Period. We only use it to route to the right models, and to fine tune models that are just yours, specialized to your unique workflows. You own those models.
- Optimization is isolated. The routing intelligence and the specialized models we spin up for your workloads are entirely isolated. They learn for you, not for us, and certainly not for your competitors.
- You control the data. We act as a strict pass-through with enterprise-grade data retention policies. When we use your data to benchmark a smaller model, it stays inside a secure, tenant-isolated environment.
Measured models, clearer choices #
Routing decisions shouldn’t be a black box, so we show our work.
See the comparisons. Our Task Explorer page shows hundreds of comparisons of various models on common enterprise workloads. You can check how each model performs before you route a single request.
Keep control. Taskrouter can handle routing automatically, but you keep ultimate control over which models are used and where your data goes.
Prove the ROI. It’s not enough to be doing AI; you have to show the business value. Taskrouter’s analytics show exactly how much money and time efficient routing saves you.
That is the whole idea: real tasks, measured models, and clearer choices.
Try it, and tell us where it breaks #
Sign up by October 10th and get $25 in free inference credits, good on any model. taskrouter.com
We also want your feedback. Throw your hardest tasks at Taskrouter and tell us where it breaks.
Stop paying frontier prices for everyday tasks. Let Taskrouter learn your workloads and optimize your AI infrastructure automatically.