Cost routing with a stated error rate

Everyone routes to the cheap model on a hunch

Short prompts to the small one, hard ones to the large one, thresholds picked in a meeting. Router picks the cheapest model that clears your quality bar with a coverage guarantee, and tells you what the guarantee costs.

No spam. One email when it is ready to try.

The problem

A routing rule with no measured error rate is just a way to lose quality quietly

Inference is the largest line on the bill, so every team writes a router. None of them can say what it costs in quality, because the rule was chosen by intuition and the failures land on the small fraction of requests nobody samples. The result is a saving you can measure and a regression you cannot.

4.5x
price gap between the cheapest and most capable model in our scenario
0.80
quality bar the mid model clears and the small one does not
none
stated error rate on a hand-written routing rule

The insight

Routing is an abstention problem wearing a cost hat

The question "is this model good enough for this request" is the same shape as "is this agent confident enough to act", and it has the same answer. Split conformal takes a calibration set and returns a lower bound on quality that holds at whatever coverage you ask for, with no assumption about the score being calibrated. Router picks the cheapest model whose bound clears your bar, so the saving comes with a number attached rather than a hope.

Method

One-sided split-conformal lower bounds per model from a graded calibration set, with the finite-sample rank correction; cheapest eligible model selected at the requested coverage, and the realised bar-hit rate checked against the promise on held-out traffic.

How it works

Four steps, no data science team

01
Grade a calibration set

A few hundred requests scored once, per model. The same set you would use for an eval.

02
State a bar and a coverage

Quality at least 0.8, at least 90% of the time. Both are business decisions, not tuning knobs.

03
Route

Cheapest model whose conformal bound clears the bar. Everything else goes up a tier.

04
Watch the promise hold

Realised hit rate against the stated coverage, continuously, so drift shows up before the bill does.

Who it is for

Whoever gets asked why the bill went up

Teams with meaningful inference spend and a quality bar they can articulate. Usually after a finance conversation about the model line.

Pricing

Free
$0
Analysis only. Tells you which model you should be using and what it would save.
  • Calibration analysis
  • Conformal bounds per model
  • Projected saving
Most common
Team
$800/mo
Live routing with continuous coverage monitoring.
  • Live routing
  • Coverage monitoring
  • Automatic re-calibration
  • Spend reporting
Scale
$3,200/mo
Self-hosted proxy, per-workload policies, enterprise controls.
  • Self-hosted proxy
  • Per-workload policy
  • SSO
  • Committed-spend reporting

Competition

What exists, and what it does not do

WhoWhat they doThe gap
OpenRouter, LiteLLM, PortkeyGateways that route across providers with fallback and caching.They route on availability, price and hand-written rules. None attaches a measured error rate to the routing decision, which is the only thing that makes a saving defensible.
Not Diamond, Martian and routing startupsLearned routers that predict the best model per request.Closest competitor, and they optimise a predicted quality score. A prediction is not a guarantee; conformal is the difference between "we think this is fine" and "this holds 90% of the time".
Prompt caching and batchingReduce cost without touching model choice.Complementary and orthogonal. Neither answers which model to use.
A rule in a config fileRoute by prompt length or customer tier.Free, already deployed, and unable to tell you what it costs in quality.
How this fails

The guarantee holds only while live traffic resembles the calibration set, and traffic mix shifts whenever the product does — so the staleness detector is load-bearing, exactly as it is for Abstain. Commercially the gateways are one feature away: adding a calibration step to LiteLLM is not hard, and they already sit in the request path with the traffic. The wedge is that doing conformal correctly and explaining why it holds is a statistics argument, and the failure mode of a naive implementation is a guarantee that quietly does not hold. That is a thin moat and an honest one.

Market

A share of the inference bill, which is the fastest-growing line in most AI budgets

Any team spending five figures a month on inference saves multiples of the subscription. A thousand teams at Team pricing is $9.6M ARR, and Scale is priced against a percentage of committed spend.

Price my routing

No spam. One email when it is ready to try.

Or just go look at the demo first →