Demo company: Corvid Cloud is fictional; the model runs are real.
$23,117$531a month.
What our Cut Your AI Bill service does, shown on Corvid Cloud’s support assistant: six changes applied one after another, with every answer scored before and after. It starts from a typical unoptimized stack: one big model on every call, a long prompt, ten retrieved passages and no length limits. Most of the saving comes from three of them: a shorter prompt and context, shorter output and easy calls sent to a cheaper model.
Quality: 76 of 81 checks pass (78 before). Every miss is shown below.
Cost per request: $0.019264 → $0.000443
Real model calls: 81 requests, scaled to 1.2M a month at the list prices of the models we ran. The service runs the same way on your bill. How we measured
As run: the bill in the headline, each call priced at the model that served it.
81 real requests, scaled to 1.2M a month, each call priced at the list price of the model that served it (Claude Sonnet 5.5, Gemini 3.1 Flash-Lite). The big models on all three price lists (Claude Sonnet 5.5, GPT-6.1 Sol, Gemini 4 Carbon) list at the same $2.00 in and $10.00 out per million tokens, so the before bill is the same on every tab; the small model's price moves the after bill. Rows are in cents so they add up to the total.
Where the money leaks
Six changes, applied one after another to the same workload. Each bar is the original bill: light grey is what is still paid after that change, green is what that change saved. Open one to see the real request behind it.
Cheaper only counts if the answers are right
Every request was scored the same way before and after: the classifier against the correct label, everything else by a grading model with a written rubric. Quality after: 76 of 81 checks pass (78 of 81 before). Failures are shown, not averaged away.
Check your own bill
Enter what you spend and how your calls look. You get a range and which changes apply; your number depends on how lean your stack already is. Nothing you type or drop here leaves your browser.
Check your bill free in your browser: the LLM Bill Checker
How we measured
One workload, run for real. These are the assumptions behind every number above.
- Scale. 81 real requests (37 support answers, 24 ticket classifications, 12 summaries, 8 nightly digests), scaled to 1.2M a month in the same mix. Tokens come from each call’s usage field.
- Prices. Each call is priced at the list price of the model that served it: Claude Sonnet 5.5 ($2.00 in, $10.00 out per million tokens) and Gemini 3.1 Flash-Lite ($0.25 in, $1.50 out), read on 11 Nov 2026. The tabs under the receipt re-price the same tokens on each provider’s list for comparison; the headline stays as run.
- Baseline. A typical unoptimized stack: one big reasoning model on every call at its default reasoning setting, a 3,073-token prompt and ten retrieved passages every time. Hidden reasoning was roughly 60% of its output tokens, estimated from the length of the visible text.
- Caching. 95% of a stable prefix of at least 1,024 tokens is served from cache. On your bill, cache storage and expiry are reconciled against the invoice.
- Quality. 76 of 81 checks pass after, 78 of 81 before. The 5 misses are listed above with the grader’s note. The first optimized pass scored 59 of 81; three fixes found by the quality check took it to 76.
- Scoring. Exact match where a correct label exists, otherwise one grading model with a written rubric, the full knowledge base and the must-have facts. On your traffic we hold back 20% of requests before tuning and score them once, at the end.
Get this on your own bill
Your bill, measured the way this page measures Corvid Cloud: a quality test from your own traffic, a ranked fix list with dollars per fix, and the code for the top fixes.
from $1,500LLM bill audit · about one week
We’re running a few free checks right now for case studies.
Cut Your AI Bill
- A quality test set built from your own traffic, so every change is scored on what your users actually send
- A ranked fix list with dollars per fix, measured on your workload
- The code changes for the top fixes, written against your codebase
- A before and after report like the full sample report built from this page’s run (PDF), or the short summary
If verified 30-day savings are below the audit fee, we refund the difference. Verified saving = (old cost per request − new cost per request) × requests served.
Start here, or go further
- Free AI check$0Send your AI bill export. You get a one-page report on your top 3 problems within 48 hours. No call needed.
- Switch Models Safely$1,500Model being retired? One model and one use case moved over. Quality on the agreed test set doesn’t drop, or we keep working free.
- AI on call$1,500/moWeekly tests on real traffic, alerts when a model update breaks something, a monthly cost and quality report. Priority $2,500/mo. Cancel any month.
Or write to hello@khlabs.dev. We reply within 24 hours. All offers