Best AI models for investment banking analysts (June 2026)

APEX: Investment Banking Analyst measures how well frontier AI models perform on real investment banking analyst tasks. Across sample tasks graded by expert banking practitioners, GPT 5.3 Codex leads at 65.0% Mean Score, ahead of GPT 5.4 (64.5%) and Gemini 3.1 Pro (63.4%).

The table below shows the top 10 AI model productivity index scores across 100 Investment Banking Analyst tasks. Scores are captured as of June 2026.

View full APEX: Investment Banking Analyst leaderboard →

Who should use APEX: Investment Banking Analyst?

The AI Model Productivity Index (APEX) for Investment Banking Analyst tasks serves 3 types of users:

• Individual investment banking analysts evaluating which AI model handles valuation, financial modeling, and quantitative analysis tasks most reliably.
• Banking technology teams standardizing AI model deployments across coverage and product groups. The benchmark gives them expert-graded performance data on the modeling tasks analysts actually do.
• Financial technology and AI companies building products for investment banking and financial analysis that need expert-graded performance data on real modeling tasks.

How are AI model productivity rankings evaluated for investment banking analysts?

These rankings come from APEX: Investment Banking Analyst, Mercor's benchmark for how frontier AI models perform on real investment banking analyst work. Unlike general reasoning benchmarks that test isolated problems, APEX: Investment Banking Analyst evaluates models on the quantitative, multi-source tasks that define analyst work, including discounted cash flow analysis, cost-of-capital calculations, and precise numerical outputs under specific rounding and formatting rules.

What does the APEX: Investment Banking Analyst benchmark measure?

APEX: Investment Banking Analyst measures frontier AI model productivity on investment banking analyst tasks across 100 prompts. Each task is passed to models with relevant source documents, mirroring how analysts actually work. Tasks are graded by human-authored rubrics and scored as Mean Score, which awards partial credit for each criterion a model satisfies.

The benchmark was built by practitioners from Goldman Sachs, Morgan Stanley, JPMorgan, Barclays, UBS, Bank of America, and Evercore.

How does APEX: Investment Banking Analyst scoring work?

Every rubric is written by an experienced banking practitioner who defines what a correct or complete financial output looks like for that task. Unlike a binary pass/fail, Mean Score awards credit for each individual rubric criterion a model satisfies. A model that correctly computes some WACC components but misses others earns partial credit reflecting the value of what it got right.

This scoring method is appropriate for financial modeling tasks where partial correctness still has value: an analysis that gets the risk-free rate and capital structure right but miscalculates beta is more useful than one that gets everything wrong, though in a live valuation a single wrong input flows through to a wrong conclusion.

What type of task is evaluated by the APEX: Investment Banking Analyst benchmark?

Here's one of the sample tasks that was evaluated within this domain:

Sample task: discounted cash flow WACC and beta calculation for Coupang

The model must perform the cost-of-capital inputs for a discounted cash flow analysis on Coupang (NYSE: CPNG). Using attached Treasury and stock-price data plus a 10-K, it must calculate the beta from 2024 daily returns against the S&P 500, determine the 2024 debt amount and 12/31/2024 market capitalization, compute the debt and equity weights, the risk-free rate, cost of equity, cost of debt, after-tax cost of debt, and the resulting WACC, all under specific rounding and units rules.

The rubric has 10 criteria. On this sample task, the top model satisfied 2 of the 10: it correctly calculated the equity to total capitalization weight and determined the risk-free rate, but missed the beta, total debt amount, market capitalization, debt weight, cost of equity, cost of debt, after-tax cost of debt, and the final WACC.

What are the hardest investment banking analyst tasks for AI models?

Multi-source financial modeling with precise numerical outputs remains the hardest category of task on APEX: Investment Banking Analyst. GPT 5.3 Codex leads at 65.0% Mean Score. The bottom of the ranked models, Opus 4.5 at 46.7% and Sonnet 4.6 at 36.9%, shows that a 28-point spread exists between the strongest and weakest models currently evaluated.

3 types of steps consistently separate stronger models from weaker ones, based on the sample task:

• Calculation from raw market data. Deriving beta by running a regression of daily returns against an index, from attached price files, requires accurate data handling that models frequently get wrong.
• Chained financial inputs. Cost of equity, cost of debt, and WACC each depend on upstream values like beta and the capital structure weights, so a single early error propagates through every downstream output.
• Precise formatting and units. Rounding percentages to one decimal, share prices to two, and presenting debt and market cap in billions to two decimals is required for each criterion, and a directionally correct answer in the wrong format still fails.

Senior analysts still outperform AI models on integrated valuation work. The judgment required to source the right inputs, run the calculations cleanly, and sanity-check the output against market reality remains largely the preserve of human expertise.

Do AI models or AI tools for investment banking matter more?

Success in investment banking depends on two things: access to the right financial information and the ability to reason through it. AI tools solve the first problem by surfacing SEC filings, market data, earnings reports, and internal research. The underlying model solves the second by turning those inputs into accurate valuations, financial models, and investment recommendations.

Better tooling cannot compensate for weak financial reasoning. A model that struggles to calculate beta from a returns series or chain together assumptions in a multi-stage valuation will still produce unreliable outputs, regardless of how much data it can access. At the same time, even the strongest model is limited without access to the filings, price histories, and rate data needed to analyze a deal.

For research summarization and market commentary, the tool often has a greater influence because it determines what information reaches the model.

For the quantitative modeling tasks measured by APEX, including multi-source valuations with precise chained inputs, the underlying model is the deciding factor. Choose the model based on the complexity of the financial analysis you need, and the tool based on how well it integrates into your deal workflow.

Get the full APEX: Investment Banking Analyst dataset

The public leaderboard shows the scores. The full dataset for investment bank analysts is what banking technology teams and financial AI companies use to make real deployment decisions, including task-level breakdowns and rubric details across all 100 tasks.