About TAB
TAB is an independent standard for measuring AI model quality.
Why TAB Exists
Most AI benchmark numbers come from the same companies that build the models. Every frontier lab is building its own evaluation inside its own walls, and a lab grading its own model is not an independent result.
TAB exists to be the opposite: a third-party measurement with no financial relationship to any model provider. It scores models on the properties that decide whether they can be trusted in production, publishes the results whether they flatter a model or not, and answers to no lab.
What Makes TAB Independent
Anthropic cannot independently verify Claude. Google cannot independently verify Gemini. OpenAI cannot grade its own homework. Independence has to come from outside the labs.
TAB writes the test cases, holds the ground truth, and scores the results through routing it controls. It does not build models. It does not sell models. It has no financial incentive for any model to score well. Nothing about a score depends on the cooperation, or the marketing, of the company being measured. That is what independent means.
How TAB Measures
TAB scores models on five properties: security, trust and reliability, agentic execution, integrity and provenance, and resilience. Every verdict is computed by fixed rules against known ground truth, never by another AI judging the output. Every run is kept as a dated, permanent record that is never overwritten, so change over time stays visible.
The full approach is documented on the methodology page.