AI spend diagnostics · report in 3 working days · from US$99
Stop paying frontier prices for routine work.
We replay your real requests against cheaper models, measure the quality gap, and show exactly which calls can move — and what you'll save.
Vendor-neutral. Fixed prices. Your data handled under NDA.
70% of traffic, moved
stay on frontier $9,000 $9,000
Assumes 70% of calls move to a model at one fifth of the price. A worked example, not a client result.
60–90%
lower prices for leading Chinese open-weight models than for top US models1
34% → 65%
open-weight share of tokens on OpenRouter, January to June 20262
3 days
from upload to a report backed by your own traffic
The problem
Token prices fall. Bills still rise.
Agents make more calls, with longer context, every quarter. Cheaper tokens don't help if you use four times as many.
Most traffic is routine — extraction, classification, summaries, simple edits — yet it runs on the most expensive model by default.
Switching endpoints takes an afternoon. Knowing what to switch, and what it does to quality, is the hard part.
How it works
Three working days. Three steps. One clear answer.
01 · Day 1
Upload
Send a usage export from your provider dashboard or gateway, plus up to 100 sample requests, anonymized if you prefer. We give you a checklist and a template.
02 · Day 2
Replay
We replay your sample against two or three cheaper candidate models — frontier lite tiers, open-weight models hosted in your region, other vendors — and score every answer against your current outputs.
03 · Day 3
Report
You get a report: where the money goes, which calls can move, what happens to quality, and what you'll save each month.
On a plan, we re-run all three steps as new models and prices land, so your setup never goes stale.
What you get
A plan your engineers can ship, not a slide of opinions.
- Spend map by task, feature and model, so you see where the money goes.
- Side-by-side evals of candidate models on your own requests.
- Savings estimate, monthly and annual, with every assumption visible.
- Your three biggest moves, ranked by savings and effort.
- On a plan, a fresh run as new models and prices land.
Vendor-neutral
We don't sell you a model.
We compare what is commercially available in your region and allowed by your procurement rules, and recommend whatever clears your quality bar at the lowest cost. If a provider ever pays us a referral fee, we tell you before you decide.
Your data
Your data stays yours.
Evals run in your environment with your own API accounts, or on an anonymized sample you approve. Before we start, we list every service that will touch your data. NDA as standard, and we delete what we hold after delivery.
Who it's for
Teams whose AI bill has outgrown its budget.
AI products and agent companies
Spending US$1k or more a month on model APIs, with costs growing faster than revenue.
Teams that must switch
Moving off a provider for policy, regional or procurement reasons, and needing a tested alternative fast.
Engineering leaders
Who want hard numbers on quality and cost before changing a model in production.
Pricing
Launch pricingStart small. Keep it current.
Snapshot
One-time · report in 3 working days
US$99
- Spend map by task and model
- Evals on up to 100 sample requests against two or three candidates
- Savings estimate and your three biggest moves
Monthly
Best while you switchFor teams actively moving workloads
US$299 / month
- A full re-run every month as new models and prices land
- Routing recommendations for each workload
- Migration questions answered by email
Annual
Quarterly check-ups for stable setups
US$599 / year
- A full re-run every quarter
- An updated savings estimate and moves each time
- Email support between check-ups
Bigger teams with many workloads get a custom scope, quoted after a short call. Hands-on implementation is quoted separately, only if you want it.
Why us
We've sat on the vendor side. Now we sit on yours.
We come from cloud vendors and strategy consulting. We know how model providers price, package and discount, and we put that knowledge on your side of the table.
Evidence, not opinion
Every recommendation traces back to evals on your own traffic.
Fixed scope, fixed price
You know the cost, timeline and deliverables before we start.
No vendor quotas
No model is ours to push. Any referral relationship is disclosed up front.
FAQ
Questions teams ask first.
Will quality drop?
We only recommend a move when your own evals show it clears your bar. Hard tasks stay where they are.
Do you need our production data?
No. Run our scripts in your environment, or share an anonymized sample you have reviewed.
Which models do you consider?
Anything commercially available in your region that your procurement and data rules allow.
What is the difference between Monthly and Annual?
Monthly is for teams in the middle of switching: a re-run every month, routing advice and email answers while you migrate. Annual is a quarterly check-up for setups that are already stable.
What if there is little to save?
Then you will know within days, for US$99, with numbers to show for it.
Are you a reseller?
No. If you want help running the new setup afterwards, we quote that separately.
Find out what your AI bill should be.
Start with a US$99 snapshot, or book a 20-minute call if you would like to talk first.
or email hello@getcounterweight.com to book a call