Independent model evaluation has become its own funded category
A crop of startups now exists solely to tell enterprises which frontier model is actually best for their workload.

As the number of credible frontier model options has grown, so has demand for independent, workload-specific evaluation — spawning a small but well-funded category of startups that do nothing but benchmark models against a customer's actual data.
The pitch is straightforward: public leaderboards measure general capability, but a legal-document summarization workload and a customer-support workload can rank different models in opposite order, and few enterprises have the tooling to test that themselves.
Several of these evaluation firms have begun publishing aggregate, anonymized findings, which vendors now watch closely — a public ranking swing has moved enterprise sales conversations within days of publication.
The category's durability is an open question. If a frontier lab's own evaluation tooling becomes good enough, the independent layer may get squeezed — but for now, enterprises are paying for the second opinion.

