AI will reshape how the world understands businesses and allocates capital. That future demands models that can follow the evidence, challenge assumptions, and exercise judgment when the answer isn’t obvious.
Our benchmarks put that intelligence to the test—through the complex, consequential work of finance.
Fifth Order Finance Index
A composite view of AI performance across financial research, modeling, and investment analysis.
Models complete multi-step financial workflows: gathering evidence, interpreting statements and turning analysis into a usable answer. Evaluation considers whether the requested work is completed, calculations reconcile and conclusions remain supported by the available information. The score summarizes task performance on a 0–100 scale; higher values indicate stronger completion across the workflow collection.
Models translate trading instructions into executable backtests, where small implementation choices can materially change the result. Evaluation checks the resulting output against the specified calculation, focusing on reliable rule interpretation and numerical agreement rather than persuasive explanations. Lower error is better. Bars use a nine-decade logarithmic scale so small errors and very large deviations remain visible together without flattening the cohort.
Models answer financial reasoning questions that require more than a plausible final number. The collection tests interpretation of financial context, selection of a defensible approach and accurate reasoning through the requested calculation. Evaluation rewards answers that reach the correct conclusion through consistent financial logic. Scores are expressed on a 0–100 scale and ranked from highest to lowest.
Models reconcile reported results and management guidance across a collection of earnings releases, filings and transcripts. Evaluation checks consistent fiscal periods and units, distinguishes actuals from forward ranges, tracks changed guidance and verifies midpoint arithmetic against cited evidence. A rubric score combines numerical reconciliation and evidence support across 40 cases; higher scores indicate more complete, defensible trackers.
Models revise investment views as new operating evidence arrives across a collection of company cases. Evaluation checks whether assumptions change with the evidence, valuation implications are explicit and counter-evidence receives fair weight. Each revision must name conditions that would invalidate the view. Scores combine these rubric criteria across 40 cases, rewarding analytical quality rather than polished prose alone.
Models are compared through one-month marked-to-market paper-portfolio total return, including uninvested cash. All ten configurations share the same observation window, September 17–October 16, 2026, with the prior close included as the starting mark. Rankings follow total return, with SPY-relative return, drawdown, volatility and exposure available in the full leaderboard.
Grok 4.6 xHigh reasoning
GPT 5.6 Luna Max reasoning
GLM 5.3 Max reasoning
