Claude & ChatGPT Compete For Wall Street

MARKETS EDITION

This week, OpenAI and Anthropic released features aimed at financial services. These two companies are the highest-earning AI model companies by annualized revenue, with OpenAI at $10B and Anthropic at $3-$4B.

We are shifting from pilots to announced partnerships and live deployments.

The largest Wall Street firms are increasingly building in-house tools, often with direct support from frontier labs like OpenAI and Anthropic. These firms are fine-tuning models or embedding them within tightly controlled internal workflows, prioritizing security, latency, and compliance.

John Hwang, founder of Enterprise AI Trends newsletter, and a veteran of Goldman Sachs, Morgan Stanley, and AWS, put it this way to me in an email.

This echoes Scale AI’s Alexandr Wang. The CEO, who just sold to 49% of his data-labeling company to Meta, says that in the AI era, intellectual property isn’t just code, but the company’s proprietary data. “Every firm’s core IP is actually their specialized model or their own fine-tuned model,” he said on a Y combinator podcast last month.

Anthropic’s Claude for Financial Services

Anthropic’s Claude for Financial Services is a suite of tools built on top of its Claude 3 models and connected to structured financial data from partners like FactSet, PitchBook, and Morningstar.

Claude for Financial Services can:

• Draft investment memos based on structured and unstructured data

• Generate and debug code for internal financial models

• Analyze and summarize earnings calls, pitch decks, and investor presentations

• Assist with portfolio monitoring and due diligence

• Access: Claude for Financial Services is available to enterprise clients.

These tools are embedded within a new architecture called Model Context Protocol (MCP), an open standard adopted by OpenAI and others, which allows companies to securely plug AI into their proprietary and third-party data systems.

Check out the demo below where an analyst leverages different data sources from the same platform.

The biggest difference from other tools seems to be how Claude handles context size and instruction following — which Anthropic says helps avoid hallucinations in financial settings.

The presentation included a discussion with executives from D. E. Shaw, HG Capital, New York Life, and Norway’s sovereign wealth fund (Norges Bank Investment Management).

They discuss practical implications such covering strategy, build vs buy decisions, culture and mindset shifts, and productivity gains.

Anthropic Adoption Examples

• Norwegian Sovereign Wealth Fund ($2T AUM): 20% productivity gains = 213,000 hours saved annually. CEO called Claude "indispensable"

• Bridgewater: Using Claude since 2023 for investment analyst assistance on complex models

• Commonwealth Bank (Australia): Making Claude the foundation of their global AI strategy

• AIG: Compressed underwriting timelines 5x (weeks to days), accuracy up from 75% to 90%

OpenAI’s ChatGPT Agent

Released Thursday, OpenAI’s ChatGPT Agent combines capabilities from two earlier tools: Operator, which could execute multi-step tasks like booking flights or sending emails, and Deep Research, which gathered and synthesized information for tasks like market analysis and benchmarking.

ChatGPT Agent goes further:

• Autonomously browse the web—including clicking, scrolling, logging in, and navigating gated sites.

• Aggregate and analyze large datasets, whether from earnings reports, regulatory filings, or internal documents.

• Generate editable spreadsheets and slide decks, enabling workflows like “compare three competitors across five KPIs” or “summarize risk disclosures from the past four quarters.”

• Streamline routine analyst tasks, OpenAI emphasized its utility for “first‑ or second‑year financial analysis” tasks that traditionally take hours.

• Access: Available to ChatGPT Plus users (GPT‑4o) and ChatGPT Team/Enterprise accounts during a phased rollout over the next couple of weeks.

The below video walks through ChatGPT agent gathering data on San Francisco’s budget from 2020 to 2024. The agent searched for the data on government websites, extracted it from reports and then displayed the budget in an excel file.

The demo claims 98% accuracy, but it’s unclear which numbers were wrong — and how a user would spot them without checking every cell.

That said, this shows that AI is getting better at handling numbers, which has been an issue for the sector. A year ago, if you put a large calculation into AI, you’d likely get a wrong answer since large language models see numbers as tokens. But now, AI is smart enough to use a calculator.

This tool clearly targets Wall Street analysts by enabling the automation of entry‑level to mid‑level financial modeling, reporting, and competitive intelligence. The chart below shows an improvement in investment banking tasks.

While this shows how AI is improving, an accuracy rate of ~71% is not going to get it done on Wall Street. You need to know how to find mistakes. That’s what makes “hallucinations” so insidious, you don’t know what’s real and what’s not.

That said, a tool that can reliably complete 70% of your work is a significant productivity boost.

Takeaway: Wall Street is moving out of the pilot phase of AI into live deployments with frontier models.

ICYMI

Check out some recent editions on AI for investment analysis*

• The Open-Source Project That’s Making SEC API Calls Cheap

• Demo: Automating Peer Groups with Primer

• Demo: Analyzing Earnings Call Tone with Markets EQ

• Demo: Separating Facts from 'AI Nonsense’ with Hudson Labs  

• Demo: AI Due Diligence with AnalystAI

• Demo: Tracking Big Tech's AI Spending with DoTadda

• Demo: Tracking DeepSeek on Earnings Calls with Aiera

• Demo: AI Agents in Market Analysis with Scalar Field

• Effective Prompts For Investment Research

*Not investment advice