How we measure.

One task, two bills, one formula, printed. How AVELIN counts cost per task, what the live Arena shows, and how you can check it.

The live Arena

In our live Arena, at list prices.

The Arena runs the same task on a flagship model and on AVELIN, then prices both. Over 122 recorded runs, 24 Sep to 2 Oct 2026, the median run cost 90.1% less on AVELIN. The 75th percentile was 94.4% and the best run 99.4%. Added up, the 122 runs cost 76.6% less.

It counts cost only. You read both answers yourself, so quality is judged separately. In 8 of the 122 runs AVELIN cost more. The comparison model differs by run, and every run is listed.

Median lower cost per task in the Arena, by the AVELIN model that answered
AVELIN modelRunsMedian lower cost per task
Chat fast4394.1%
Chat pro5492.5%
Chat ultra1762.8%
Coding pro633.8%
Agents2Too few runs to report

The Arena is a public demo with its own privacy page, so do not paste confidential data into it.

Cost per task

Cost per task is the sum of every model call a task makes, each priced at list price. Tokens in and tokens out are multiplied by the published rate for the model that answered. Cached tokens are priced at the cache rate, and reasoning tokens are billed as output. AVELIN list prices are on the price list, as of 3 Oct 2026. We use no negotiated prices, volume discounts or credits, on either side.

cost = (uncached input tokens × input price + cache read tokens × cache read price + cache write tokens × cache write price + output tokens × output price) ÷ 1,000,000

The Decision Layer

AVELIN's own trained decision model reads each request and picks the model it needs: a more capable model when the task needs intelligence, a cheaper one when it is easy, on every request.

  • Reads. The opening of a request, about the first 1,500 characters.
  • Predicts. The kind of job, from 15 categories, and how hard it is, on a 5-level scale.
  • Chooses. The model that fits, a more capable one for hard tasks and a cheaper one for easy tasks. A task of several steps can change model between steps.
  • Records. Every decision is logged with its reason, so a choice can be checked afterwards against what each model would have cost.

Check it yourself

Price the AVELIN side of any task with the formula above. Send a prompt to the API, read the token counts in the response and multiply them by the list prices. Run the same prompt on the model you use today, then compare the two answers and the two bills.

curl https://api.avelin.ai/v1/chat/completions \  -H "Authorization: Bearer $AVELIN_API_KEY" \  -H "Content-Type: application/json" \  -d '{"model": "avelin-pro", "messages": [{"role": "user", "content": "Your prompt here"}]}'

In a POC the other side is your own bill, and the report is your own workload at list prices.

Earlier sample: 60 runs on 26 Sep 2026

On 26 Sep 2026 we ran 15 sample tasks, each once against 4 flagship models: 60 runs, with both bills at list prices. The prompts and every AVELIN answer, as recorded, are in one JSON file.

Where AVELIN cost more

The recorded runs where the AVELIN bill was higher
RunTaskFlagshipAVELIN tierAVELIN billFlagship bill
1GDPR: a hospital sends patient records to a US AI vendorLegal, hardCavelin-ultra$0.10193$0.06337
2GDPR: a hospital sends patient records to a US AI vendorLegal, hardDavelin-ultra$0.09424$0.04616
All 60 recorded runs
All recorded runs, losses first
RunTaskFlagshipAVELIN tierAVELIN billFlagship bill
1GDPR: a hospital sends patient records to a US AI vendorLegal, hardCavelin-ultra$0.10193$0.06337
2GDPR: a hospital sends patient records to a US AI vendorLegal, hardDavelin-ultra$0.09424$0.04616
3Explain an overdraft fee notice to a customerBanking, easyAavelin-fast$0.00030$0.00763
4Explain an overdraft fee notice to a customerBanking, easyBavelin-fast$0.00030$0.01097
5Explain an overdraft fee notice to a customerBanking, easyCavelin-fast$0.00030$0.01049
6Explain an overdraft fee notice to a customerBanking, easyDavelin-fast$0.00618$0.03725
7Flag suspicious transactions for an AML reviewBanking, mediumAavelin-pro$0.00256$0.03922
8Flag suspicious transactions for an AML reviewBanking, mediumBavelin-pro$0.00285$0.05584
9Flag suspicious transactions for an AML reviewBanking, mediumCavelin-pro$0.00243$0.01388
10Flag suspicious transactions for an AML reviewBanking, mediumDavelin-pro$0.00255$0.03229
11Credit decision: DSCR, leverage and collateral for a business loanBanking, hardAavelin-pro$0.00605$0.10047
12Credit decision: DSCR, leverage and collateral for a business loanBanking, hardBavelin-pro$0.00794$0.09895
13Credit decision: DSCR, leverage and collateral for a business loanBanking, hardCavelin-pro$0.00638$0.11715
14Credit decision: DSCR, leverage and collateral for a business loanBanking, hardDavelin-pro$0.00706$0.10886
15Fresh facts: today's ECB deposit facility rate and its last changeBanking, mediumAavelin-fast$0.01272$0.04115
16Fresh facts: today's ECB deposit facility rate and its last changeBanking, mediumBavelin-fast$0.01245$0.08149
17Fresh facts: today's ECB deposit facility rate and its last changeBanking, mediumCavelin-fast$0.01847$0.03990
18Fresh facts: today's ECB deposit facility rate and its last changeBanking, mediumDavelin-fast$0.01842$0.04385
19Rewrite discharge instructions for the patientHealthcare, easyAavelin-pro$0.00111$0.01785
20Rewrite discharge instructions for the patientHealthcare, easyBavelin-pro$0.00918$0.04129
21Rewrite discharge instructions for the patientHealthcare, easyCavelin-pro$0.00108$0.01412
22Rewrite discharge instructions for the patientHealthcare, easyDavelin-pro$0.00101$0.05484
23Turn a clinical note into structured JSONHealthcare, mediumAavelin-pro$0.00207$0.02760
24Turn a clinical note into structured JSONHealthcare, mediumBavelin-fast$0.00075$0.02393
25Turn a clinical note into structured JSONHealthcare, mediumCavelin-pro$0.00991$0.01260
26Turn a clinical note into structured JSONHealthcare, mediumDavelin-fast$0.00666$0.02834
27Medication reconciliation with poor kidney functionHealthcare, hardAavelin-ultra$0.03754$0.17942
28Medication reconciliation with poor kidney functionHealthcare, hardBavelin-pro$0.01165$0.14904
29Medication reconciliation with poor kidney functionHealthcare, hardCavelin-ultra$0.05371$0.10061
30Medication reconciliation with poor kidney functionHealthcare, hardDavelin-pro$0.01064$0.06347
31Explain an NDA clause in plain EnglishLegal, easyAavelin-pro$0.00113$0.00955
32Explain an NDA clause in plain EnglishLegal, easyBavelin-pro$0.00110$0.01845
33Explain an NDA clause in plain EnglishLegal, easyCavelin-pro$0.00105$0.00775
34Explain an NDA clause in plain EnglishLegal, easyDavelin-pro$0.00108$0.01634
35Compare two liability clauses: what changed, who benefitsLegal, mediumAavelin-pro$0.00412$0.05856
36Compare two liability clauses: what changed, who benefitsLegal, mediumBavelin-pro$0.00392$0.10492
37Compare two liability clauses: what changed, who benefitsLegal, mediumCavelin-pro$0.00553$0.01282
38Compare two liability clauses: what changed, who benefitsLegal, mediumDavelin-pro$0.00476$0.02836
39GDPR: a hospital sends patient records to a US AI vendorLegal, hardAavelin-ultra$0.16501$0.22871
40GDPR: a hospital sends patient records to a US AI vendorLegal, hardBavelin-ultra$0.08758$0.36907
41Fresh facts: the EU AI Act duties that already apply to GPAI providersLegal, mediumAavelin-pro$0.03017$0.13797
42Fresh facts: the EU AI Act duties that already apply to GPAI providersLegal, mediumBavelin-pro$0.03036$0.22198
43Fresh facts: the EU AI Act duties that already apply to GPAI providersLegal, mediumCavelin-pro$0.03027$0.05687
44Fresh facts: the EU AI Act duties that already apply to GPAI providersLegal, mediumDavelin-pro$0.03052$0.08350
45Draft a reply to this customer complaintGeneral, easyAavelin-pro$0.00096$0.00934
46Draft a reply to this customer complaintGeneral, easyBavelin-pro$0.00096$0.02518
47Draft a reply to this customer complaintGeneral, easyCavelin-pro$0.00085$0.00778
48Draft a reply to this customer complaintGeneral, easyDavelin-pro$0.00076$0.02302
49Extract the fields from this invoiceGeneral, easyAavelin-fast$0.00031$0.00626
50Extract the fields from this invoiceGeneral, easyBavelin-fast$0.00031$0.00820
51Extract the fields from this invoiceGeneral, easyCavelin-fast$0.00031$0.01675
52Extract the fields from this invoiceGeneral, easyDavelin-pro$0.00059$0.01372
53Summarise this contract clause for the CFOGeneral, mediumAavelin-pro$0.00088$0.01449
54Summarise this contract clause for the CFOGeneral, mediumBavelin-pro$0.00100$0.02420
55Summarise this contract clause for the CFOGeneral, mediumCavelin-pro$0.00105$0.00753
56Summarise this contract clause for the CFOGeneral, mediumDavelin-pro$0.00113$0.01620
57Write the SQL for last month's churn by segmentGeneral, mediumAavelin-coding-pro$0.00767$0.03360
58Write the SQL for last month's churn by segmentGeneral, mediumBavelin-coding-pro$0.01306$0.04848
59Write the SQL for last month's churn by segmentGeneral, mediumCavelin-coding-pro$0.01659$0.06667
60Write the SQL for last month's churn by segmentGeneral, mediumDavelin-coding-pro$0.03075$0.06307

Download the 60 runs as a CSV file. It has every pair of bills, the search fee and the step bills.

Questions about the method

How is cost per task counted?

Cost per task is the sum of every model call a task makes, each priced at list price. Tokens in and out are multiplied by the published rate for the model that answered, cached tokens are priced at the cache rate, and reasoning tokens are billed as output.

What does the live Arena measure?

In our live Arena, at list prices. The Arena runs the same task on a flagship model and on AVELIN and prices both. Over 122 recorded runs from 24 Sep to 2 Oct 2026, the median run cost 90.1% less. It counts cost only, and quality is judged separately: you read both answers yourself.

What does the decision model read and predict?

It reads the opening of a request, about the first 1,500 characters. It predicts the kind of job, from 15 categories, and how hard it is, on a 5-level scale. The Decision Layer then picks the model that fits and logs the reason.

Can I check the figures myself?

Yes. Every Arena run is listed in a public CSV and on the Arena scoreboard. You can also send a prompt to the API, read the token counts in the response and apply the formula on this page.

How does a POC measure cost?

A POC runs the Decision Layer on a sample of your own tasks and prices each task at list price next to your current bill. The report lists the tasks where the Decision Layer costs more first.

Request a POC

Measure it on your own tasks.

A POC runs the Decision Layer on a sample of your tasks and prices each one next to your current bill.