Open-weight model assurance
Is this model safe and secure enough for your use case?
DuraGym tests open-weight models against a catalog of benchmarks and your own acceptance criteria, shows why a model fails where it fails, and ends in a release and a report a risk owner can sign off on.
- 01Modelspinned and scanned
- 02Benchmarksa versioned catalog
- 03Evaluategates you set
- 04Trainfix what fails
- 05Releasesigned evidence
Sign in
Continue with an approved Google or GitHub account.
Access is limited to approved accounts. To ask for access, write to Dura Labs.