A new AI every month week day.

Own your AI evals.

No benchmaxxing, no drama.
Your work, tested on every new AI.

Examples invoice due dates drug names in Mandarin dictation damaged pallets support reply tone

formosa-rx/drug-names-zh-tw Example eval

Drug names in Mandarin dictation

Three models are too close to call.

  1. 1–3gpt-6.1-sol-pro91.4 ±2.9
  2. 1–3scribe-v390.8 ±3.0
  3. 1–4gemini-3.8-flash89.6 ±3.2
  4. 2–6gpt-6.1-sol87.9 ±3.4
  5. 4–8claude-opus-5.585.3 ±3.7

270 private clips from 14 pharmacists. Whiskers show the 95% range.

Checked by breakfast.

Your evals Best before This model Against best

Open an eval to read its graded answers.

No peeking.

Priced to cover servers, not to profit.

Plan

Self-host

$0

Open source, on your servers.

Get the code

Cloud

$0

We host it. You pay for model calls.

Create eval

Supporter

$24per year

More storage and every answer kept.

Support us
Public and private evals
Unlimited
Unlimited
Unlimited
Storage
Your disk
10 GBOr your Hugging Face account
100 GB
Checks
Weekly and new models
Weekly and new models
Weekly and new models
Eval history
Forever
ForeverAnswers kept 6 months
Forever
Team members
Yes
Yes
Yes
Local models and GPUs
Yes
No
No

Models cost their list price, paid from credits or your own key.

Write one sentence tonight.

The owl will have results by breakfast.

Create eval