Back to the project

What smart routing saves

Agentlify’s cost and quality results, in production and on a public benchmark.

Most teams send every request to their strongest, most expensive model. Agentlify routes each request to the model that gives the best quality per dollar for that kind of task, and keeps learning from every eval result.

01

What does Agentlify save in production?

Half the inference bill, for about a 5% drop in eval score, on real customer traffic.

50%

Lower inference cost

average across production traffic

~5%

Eval score change

versus always using the strongest model

Source: Agentlify production dashboard.

How each request is routed

  1. Step 1

    Embed

    The request is turned into a vector that captures what it asks for.

  2. Step 2

    Match

    It is matched to clusters of similar past requests.

  3. Step 3

    Sample

    A Thompson-sampling bandit picks the model with the best expected quality per dollar for that cluster.

  4. Step 4

    Learn

    The eval result updates that model’s record for the cluster, so the next pick is better.

02

Does it hold on a public benchmark?

On RouterBench, with 11 real models and their real per-request prices, Agentlify cuts cost by 79% while staying within 5% of GPT-4’s score. A standard classifier router gets 54%, and simply mixing single models gets 28%.

Agentlify

79%

cost saved within 5% of GPT-4’s score

95% CI 77–81%

Classifier router

54%

cost saved within 5% of GPT-4’s score

95% CI 41–64%

Agentlify Classifier router Zero router (best mix of single models) Single-model baselines Within 5% of GPT-4’s score

↑ Eval score vs always using GPT-4

−20%−15%−10%0%−5%
54%
79%
Always Yi-34B
0%25%50%75%100%

Cost saved vs always using GPT-4 →

RouterBench 0-shot, 11 models, 7,300 test prompts, the dataset’s real per-request costs. Each line sweeps how much the router weighs cost against quality. Mean of 20 random orderings of the test stream. Off the scale: perfect routing (the oracle) would score +17.0% while saving 93%; picking a model at random scores −33.1%.

Show as table
RouterCost weightCost savedEval changeGPT-4 calls
Agentlify019.4%+0.7%74.5%
Agentlify0.0122.5%+0.7%70.9%
Agentlify0.0223.8%+0.7%69.2%
Agentlify0.0324.6%+0.7%68.6%
Agentlify0.0426.7%+0.7%67.8%
Agentlify0.0545.8%+0.3%65.4%
Agentlify0.0666.1%−0.3%61.7%
Agentlify0.0870.8%−1.0%56.3%
Agentlify0.173.9%−2.3%49.1%
Agentlify0.1277.9%−4.4%37.8%
Agentlify0.1588.3%−9.0%11.1%
Agentlify0.291.1%−10.6%4.2%
Agentlify0.2593.1%−11.8%0.0%
Agentlify0.393.1%−11.9%0.0%
Agentlify0.493.2%−11.8%0.0%
Agentlify0.593.2%−11.8%0.0%
Classifier router05.2%+0.4%93.4%
Classifier router0.016.4%+0.6%91.7%
Classifier router0.027.8%+0.4%90.0%
Classifier router0.039.7%+0.3%87.6%
Classifier router0.0411.6%+0.1%84.6%
Classifier router0.0514.3%−0.2%81.4%
Classifier router0.0618.0%−0.7%77.1%
Classifier router0.0823.0%−1.8%67.6%
Classifier router0.128.4%−3.2%55.6%
Classifier router0.1259.1%−5.4%34.0%
Classifier router0.1585.5%−8.3%14.5%
Classifier router0.291.5%−10.8%4.3%
Classifier router0.2593.3%−12.3%0.1%
Classifier router0.393.4%−12.4%0.0%
Classifier router0.493.5%−12.5%0.0%
Classifier router0.593.5%−12.5%0.0%
Always GPT-4–0.0%0.0%–
Always Yi-34B–94.4%−17.1%–
Always cheapest (Mistral-7B)–98.6%−60.3%–
Random model–74.7%−33.1%–
Oracle (perfect routing)–92.6%+17.0%–

03

What does that mean in dollars?

The same results as a monthly bill. Enter what you spend on models today.

Based on

You’d save

$5,000

a month · $60,000 a year

Today · every request to your strongest model$10,000
With Agentlify · 50% lower$5,000

50% lower cost for about a 5% lower eval score, measured on production traffic. Your savings depend on your traffic mix and which models you use.

04

What if a model gets worse?

Model quality changes with every provider update. In simulation, a router trained once never recovers when the ranking flips. The bandit router recovers on its own, and with forgetting, which down-weights old results, it loses about half as much quality as a router retrained every 1,000 requests.

Simulated: synthetic models, not real LLMs or production traffic.

Bandit router, with forgetting Bandit router, default settings Static router, retrained every 1,000 requests Static router, trained once

↑ Quality of the chosen model (rolling average of 100 requests)

0.60.70.8ranking reverses
01,0002,0003,0004,0005,000

Requests →

Quality lost over 5,000 requests

Cumulative regret against the best model per request; lower is better. Thin bars are 95% intervals.

Bandit router, with forgetting
110
Bandit router, default settings
288
Static router, retrained every 1,000 requests
204
Static router, trained once
596

When nothing changes, forgetting costs little: 71 against 66 with default settings, over the same 5,000 requests. 3 synthetic models, 5 request clusters, 20 seeds.

Show as table
RouterRegret (drift)95% CI
Bandit router, with forgetting109.6103.4 to 115.8
Bandit router, default settings287.7266.6 to 308.7
Static router, retrained every 1,000 requests203.9180.9 to 227.0
Static router, trained once596.1530.9 to 661.3

Agentlify is at agentlify.co.