FrontierFinance (Financial Modeling) - Three-Statement: leaderboard

FrontierFinance (Krumdick et al., 2026; distinct from the Samaya AI FrontierFinance board) gives computer-use agents long-horizon financial modeling tasks: from a detailed prompt the agent builds a complete Excel model of a real company, scored against an expert-written rubric. Average judge rubric score over the five three-statement model tasks, in percent; higher is better.

Metric: Average rubric score (%) over the five three-statement model tasks, judged by a Claude Sonnet 4.6 agent against the expert-written task rubric (with SEC EDGAR and spreadsheet tools); computer-use agents in a minimal Inspect harness (GUI control plus a bash terminal) build complete Excel financial models from detailed prompts; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.4xHigh76.9#135
2Claude Opus 4.6High76.2#113
3Gemini 3.1 Pro (Preview)54.9#99

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=frontierfinance-financial-modeling-three-statement · How It Works · Data refreshed daily, snapshot 2026-11-21.