Why eval startups fail (2025)

Independent eval startups consistently fail because top talent gravitates toward more lucrative and prestigious roles in post-training and application development rather than staying in the eval niche, the addressable market is tiny since developers skilled enough to need evals are usually capable of building their own, and eval startups face optimization pressure from clients and model developers that quickly renders their benchmarks obsolete. The only exception is safety-focused eval companies, which operate in a different market dynamic than commercial eval vendors.