Thrive Holdings publishes its first real-world AI benchmarks for tax and IT work

Thrive Holdings publishes its first set of AI model benchmarks drawn from work across its more than 75 accounting and IT services firms, at evals.thriveholdings.com.

The company says it uses the tests to compare model configurations and choose which models its systems run on, with the goal of better service to customers. Thrive Holdings says the field needs more benchmarks tied to economically valuable work so builders can see where models are useful in real environments. OpenAI's Boris Power praised the release on X.