Vals Finance Index finds no clear model winner across research, Excel and tax
The composite benchmark puts four systems within one point at the top while exposing large cost differences and several methodological caveats.
2 min read
Vals AI has introduced the Finance Index, a composite leaderboard intended to compare models across three kinds of finance work: research, spreadsheet modeling and tax questions.
The index combines Finance Agent Benchmark v2.1, Excel Modeling Benchmark v1.1 and Tax Agent Bench. Only systems evaluated on all three are ranked. Vals weights the components using U.S. Bureau of Economic Analysis GDP figures: securities and investment activity receives a 1.6 weight, while tax preparation receives 0.5.
As of October 21, Vals placed Claude Opus 5.5 first at 65.59%, followed by Gemini 4 Argon at 64.89%, Wonder Full at 64.55% and Claude Sonnet 5.5 at 64.53%. Vals explicitly says the leading systems are within the benchmark’s margin of error, so the ranking does not establish a clear winner. No model led every component.
Cost also changes the picture. The leaderboard lists average costs of $11.67 per test for Opus, $6.53 for Gemini 4 Argon, $3.14 for Wonder Full and $8.33 for Sonnet. DeepSeek v4.1 Flash 1017 scored 60.89% at $0.42 per test. Two Anthropic entries include some fallback-model completions after refusals, another reason to avoid treating the scores as a perfectly controlled head-to-head comparison.
The Finance Index is proprietary, and its results have not been independently replicated. Its value is therefore in the transparent component weighting and cost context—not in declaring a universal “best model” for finance.
Image credit: Photo by Rodrigo Rodrigues | WOLF Λ R T on Unsplash, used under the Unsplash License.
Sources: Vals Finance Index and methodology · Image · Unsplash License
Conversation
0 approved commentsNo comments yet. Start a thoughtful conversation.
Join the conversation.
Sign in with ChatGPT to leave a comment or rate this article.