Numa 1

Every benchmark where Numa 1 has a score, best placing first. Click a benchmark for its full leaderboard.

About

Numa 1 is the model behind Moneda AI, the agent in the Moneda app. It researches stocks, ETFs and companies with live market data and reads the filings itself. Scores here are from runs Moneda did on the product as users get it.

moneda.com/models/numa-1

Numa 1

Moneda

Benchmarks scored6

Finance Agent BenchmarkAnalyst Financial research · Partial Credit accuracy % · 9 Oct 2026: 80.2#1 of 11
FinanceBenchFinancial filings Q&A · % correct, no filing given (150-question sample) · 2 Nov 2026: 67#1 of 5
FinSearchCompLive financial market data · Global subset average accuracy % · 1 Nov 2026: 59.4#3 of 11
FinanceQAFinancial filings Q&A · accuracy (total, 0-1) · 1 Nov 2026: 0.355#4 of 5
Moneda FilingsBenchFinancial filings Q&A · % correct (private split) · 2 Nov 2026: 91.6#1 of 1
Moneda PortfolioBenchFinancial trading and portfolios · Portfolio tasks, % correct · 1 Nov 2026: 96#1 of 1

By question type

From Moneda's runs of 2026-11-01. Leaderboards rank on each benchmark's overall score; this shows where that score comes from.

FinSearchComp

Simple lookups of past figures, Greater China (100 questions): 73.0%
Simple lookups of past figures, global (119 questions): 67.6%
Complex multi-step investigations, global (83 questions): 51.2%
Complex multi-step investigations, Greater China (86 questions): 41.3%

FinanceBench

Standard metrics from filings (50 questions): 91.0%
Questions needing finance judgement (50 questions): 57.0%
Newly written questions (50 questions): 53.0%

FinanceQA

Conceptual (64 questions): 53.9%
Basic analyst calculations (38 questions): 36.8%
Stating and using your own assumptions (46 questions): 8.7%

Moneda FilingsBench

Fourth-quarter figures (50 questions): 100.0%
Reportable segments (13 questions): 97.4%
Fiscal years not ending in December (37 questions): 97.3%
Multi-step calculations (58 questions): 96.6%
Comparisons (17 questions): 94.1%
Several companies (44 questions): 90.9%
Figures as first filed, before restatement (68 questions): 88.2%
Simple lookups (24 questions): 87.5%
Ratios with the formula given (10 questions): 80.0%
Notes to the statements (40 questions): 77.5%