How The Fb Composite Score Is Calculated - FinanceBenchmark Research

How the FB Composite Score is calculated

A plain-language explanation of the weighted average behind the FB Score - what the tier weights mean, why we normalise by evaluated benchmarks rather than total benchmarks, and what the asterisk means.

FinanceBenchmark Research

financebenchmark.ai

·Published Jun 25, 2026·Updated Jun 25, 2026·5 min read

The basic idea

The FB Score is a weighted average. We take each benchmark score, multiply it by that benchmark's weight, sum the results, and divide by the total weight of the benchmarks that model has been evaluated on. That is it. The complexity is in the weights - which benchmarks count more, and why.

Tier weights

We assign every benchmark to one of three tiers based on task complexity and the directness of financial relevance. Tier 1 benchmarks - PAVE and Finance Agent - get a weight of 30. Tier 2 - FinanceBench, FinQA, ConvFinQA - get 20. Tier 3 - FinBen, MultiHiertt, FinAuditing - get 10. The intuition is that a model which can verify an interchange fee from first principles is demonstrating a harder capability than one that can retrieve a revenue figure from a filing.

Why we normalise

We divide by the sum of weights for benchmarks that model has been evaluated on - not by the total weight of all eight benchmarks. This means a model evaluated only on PAVE has an FB Score equal to its PAVE score, rather than being penalised for the seven benchmarks it has not been run on. The risk is that single-benchmark scores look comparable to multi-benchmark composite scores. We address this with the asterisk.

What the asterisk means

Any model evaluated on fewer than three benchmarks displays an asterisk next to its FB Score. This is a signal, not a demerit. It means the composite score is based on limited coverage and should be interpreted as a partial view of the model's financial reasoning capability. We include partial evaluations rather than excluding them - see the methodology page for why.