ChatGPT vs Claude for Financial Analysis

Claude has the edge for financial analysis on the most-cited public benchmark and across Microsoft Office: on Vals AI’s Finance Agent leaderboard updated October 29, 2026, Claude Opus 5.5 ranks first at 67.85%, against 61.82% for OpenAI’s highest-scoring model on the leaderboard, GPT-6.1 Sol, and Claude’s add-ins cover Excel, PowerPoint, Word and Outlook. ChatGPT for Financial Services has the edge on included data, with SEC filings, transcripts and financial datasets in the plan without separate contracts. For numbers a firm will act on, the stronger choice is the Kepler platform, which runs inside either assistant and computes every figure in code, cited to the exact page, line or cell.

Both are excellent frontier assistants. Both, used on their own, still leave the analyst checking the numbers before they go into a model or a memo, and that’s the work the Kepler platform takes on from inside either one.

We build Kepler, which runs inside both ChatGPT and Claude. Anthropic has published a case study on how we built Kepler with Claude, and inside Kepler we route each task to whichever model does it best. Facts below come from Anthropic, OpenAI and Vals AI as of November 2026, linked inline.

ChatGPT vs Claude at a glance

	ChatGPT for Financial Services	Claude for Financial Services
Launched	September 10, 2026	July 15, 2025
Model	GPT-6 Astra, with newer models added as released	Claude Opus, Sonnet and Haiku models
Who can buy it	Eligible financial institutions, as a separate plan on ChatGPT Enterprise	Banks, insurers, asset and wealth managers, fintechs
Financial data	Included: SEC filings, Quartr, Daloopa, Fiscal.ai, PitchBook Essentials, Crunchbase, LSEG News (US), delayed pricing	Connectors to LSEG, FactSet, S&P Capital IQ, Morningstar, PitchBook, Moody’s, Dun & Bradstreet, Daloopa, through your licenses
Other connectors	50+, including Datasite, Box, Preqin, FactSet, Intapp	Box, Databricks, Snowflake, your own MCP servers
Office	ChatGPT for Excel and Google Sheets; firm Excel, Word and PowerPoint templates	Claude for Excel, PowerPoint and Word; Outlook in beta
Citations	Figures traced to tables and passages, highlighted	Figures and claims linked to underlying data and documents
Vals AI Finance Agent v2.1	GPT-6.1 Sol 61.82%; GPT-6 Astra 60.43%	Claude Opus 5.5 67.85%; Claude Sonnet 5.5 67.28%
Training on your data	Not by default	Not by default on Team and Enterprise
Kepler inside	Yes	Yes, through a verified MCP connector

The two finance offerings

ChatGPT for Financial Services

OpenAI introduced ChatGPT for Financial Services on September 10, 2026. It’s a separate plan built on ChatGPT Enterprise, sold to eligible financial institutions, and it starts with investment banking and equity research. It runs on GPT-6 Astra, and OpenAI says newer models will be available out of the box as they’re released.

The headline feature is data OpenAI indexes and hosts itself, so teams can start without negotiating contracts or setting up connectors. Its help page lists SEC filings; Quartr transcripts, investor presentations and international filings; Daloopa financials with a 24-hour delay and a 3,000 data point monthly cap per user; Fiscal.ai fundamentals; PitchBook Essentials; Crunchbase; LSEG News for US-based users; and US equity pricing from FMP and Nasdaq, delayed 15 minutes. Citations open the specific table or passage with the evidence highlighted. Admins can publish firm Excel, Word and PowerPoint templates so research notes, models and pitchbooks come out in house format.

Claude for Financial Services

Anthropic launched Claude for Financial Services on July 15, 2025, and has expanded it since. Anthropic’s solutions page describes one package of models, apps, data connectors and implementation support for banks, insurers, asset and wealth managers and fintech firms. Data comes through pre-built connectors to LSEG, FactSet, S&P Global (Capital IQ), Morningstar, PitchBook, Moody’s, Dun & Bradstreet and Daloopa, and firm data through Box, Databricks and Snowflake. Firms can deploy through Claude Enterprise, Claude for Microsoft 365, Claude Code or the API, on AWS, Google Cloud or Azure, with a dedicated financial services team and consulting partners for rollout.

Data: included or connected

The two companies made different bets.

OpenAI bundles. Datasets arrive with the plan and work on day one, which suits a team that doesn’t hold a FactSet or Capital IQ license. The trade-off is depth: OpenAI says the included datasets provide “selected coverage, which may differ from the provider’s full product.” It’s also working on shared sign-in with S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva and Moody’s, so users can reach data they’re already entitled to.

Anthropic connects. Claude reaches the providers your firm already pays for, at the depth of your license, which suits firms with established data contracts. The trade-off is setup: each premium connector needs your own subscription with that provider.

Neither one knows your firm’s internal definitions unless you supply them. Both learn what “adjusted EBITDA” means at your shop from what you connect, upload or template.

Excel, PowerPoint and Word

ChatGPT for Excel launched in beta on March 5, 2026 and became generally available on all plans on May 5, running GPT-5.5, with a Google Sheets version alongside. It builds and updates models in plain language, explains how assumptions flow through a workbook, links answers to the cells it references and asks permission before changing anything. On OpenAI’s own investment banking benchmark, which includes building a three-statement model with formatting and citations, OpenAI reported GPT-5.4 Thinking at 87.3%, up from 43.7% for GPT-5.

Claude for Microsoft 365 puts Claude in Excel, PowerPoint and Word, generally available on all paid plans, with Outlook in beta. Context carries across apps, so a brief in Word can become a model in Excel and then a deck in PowerPoint without re-explaining the work. Claude follows the workbook’s formula conventions and the deck’s slide masters. It’s also available through Amazon Bedrock, Gemini Enterprise Agent Platform and Microsoft Foundry.

Microsoft’s own Agent Mode in Excel lets eligible Copilot users pick OpenAI or Anthropic models inside Excel, so some firms reach both without choosing.

Benchmarks: what Vals AI’s Finance Agent shows

Vals AI’s Finance Agent benchmark tests whether a model can do the work of an entry-level analyst: answering hard questions from public company filings, with tools to search EDGAR and read documents. Version 2.1, updated October 29, 2026, scores models on a private test set of 449 questions, graded by a jury of three frontier models. Answers that hedge between two values now fail.

Rank	Model	Lab	Accuracy
1	Claude Opus 5.5	Anthropic	67.85%
2	Gemini 4 Argon	Google	67.66%
3	Wonder Full	Meta	67.58%
4	Claude Sonnet 5.5	Anthropic	67.28%
5	Gemini 4 Carbon	Google	67.16%
9	Claude Haiku 5.5	Anthropic	65.37%
14	GPT-6.1 Sol	OpenAI	61.82%
22	GPT-6 Astra	OpenAI	60.43%

The table lists the leaderboard’s top five plus the other Anthropic and OpenAI models discussed here, with Vals AI’s ranks. Three details matter for analysts. Anthropic’s models hold three of the top nine places. The best score in the Financial Modeling category is 57.0%. And Vals notes that when only fully correct answers count, scores “fall substantially across the leaderboard.” The leader earns about two-thirds of the available credit on work Vals describes as entry level, so any figure either assistant produces still needs a source and a second look.

Security and governance

ChatGPT for Financial Services builds on ChatGPT Enterprise: SAML SSO, SCIM provisioning and role-based access. OpenAI says business data isn’t used for training by default, data is encrypted at rest and in transit, and admins set retention. Compliance teams can export workspace logs through the OpenAI Compliance Platform, and firms can create multiple workspaces to enforce information barriers. Standard Enterprise seats and Financial Services seats can’t share a workspace.

Claude for Financial Services comes with Anthropic’s security documentation, published on its trust center. Anthropic doesn’t train on Team or Enterprise data by default, and firms can run Claude through AWS, Google Cloud or Azure.

Which should you choose?

ChatGPT for Financial Services fits when you want financial data included without negotiating contracts, your work is investment banking or equity research, your firm has standardized on OpenAI, or your team works in Google Sheets.

Claude for Financial Services fits when your firm already licenses FactSet, Capital IQ, LSEG or PitchBook, your work moves across Excel, PowerPoint, Word and Outlook, benchmark accuracy weighs heavily, or your engineers build on the API and Claude Code.

Either way, both vendors tell users to check the numbers. OpenAI’s help page says to “check important information against the supporting sources” and “review outputs before using them in client materials or investment decisions.” Anthropic says Claude links figures to underlying data “so teams can verify work before they act.” Whichever one you pick, analysts can call Kepler from inside it, so the numbers that go into a model come back computed in code and cited to the line.

What neither does on its own

A frontier model on its own writes its figures as text, so a number can be wrong without looking wrong, and a citation can show that a source exists without proving the figure came from it. It reads the passages it retrieves, so a question about every fund agreement with a key-person clause comes back as the top matches. It learns your firm’s definitions from whatever you paste or connect. And asked the same question twice, it can return two different numbers. Our guide Can ChatGPT do financial analysis? walks through each of these, AI hallucinations in finance explains why better models make the errors quieter rather than gone, and how to check AI-generated numbers gives the tests.

The third option: Kepler

We build Kepler.

Kepler works alongside whichever assistant your analysts prefer. It runs inside ChatGPT and Claude through MCP (how MCP works for financial data), so when a question touches your firm’s data or a number someone will act on, the answer comes from the Kepler platform.

Three things change. First, our forward-deployed engineers build your firm’s ontology, one living model of your funds, holdings, companies, managers and house definitions, kept current as documents arrive and able to answer as of any date. Second, extraction models we train read every page before anyone asks: 26M+ SEC filings across 20+ form types, 40M+ transcripts and IR documents, and everything your firm connects from SharePoint, OneDrive, Box, Egnyte, Outlook, Slack and your CRM. Those models are small: one cost us $9 to train. Third, code computes every figure. Statements assemble from XBRL as filed or as restated, calculations run over cited inputs with their formulas shown, every number opens on the exact page, line or cell, and citations are checked against the source before the answer reaches you.

Your firm’s templates are saved once as playbooks, so a Kepler answer arrives as the comps sheet, model or note your team already uses. Behind each answer, we route every step to the model that does it best and swap in better ones as they ship, so each new frontier model, from either lab, makes Kepler better. Agents watch filings and news on a schedule and act as the person who set them up. Every read passes one permission check, every query is logged, nothing Kepler reads is ever taken as an instruction, and we hold SOC 2 Type II. Our engineers built this layer before at Palantir, and Anthropic’s case study on Kepler describes one of our models scoring 94% accuracy on mapping financial statement labels to standardized taxonomy codes, against 38% to 46% for other models. For a step-by-step setup in Claude, see how to use Claude for financial analysis.

Questions analysts ask Kepler from inside ChatGPT or Claude:

• “EBITDA margin by segment for our coverage list over 12 quarters, on our adjusted definition, with every input cited.”
• “Which holdings changed full-year guidance this week? Quote the exact sentence.”

Not the right fit if: you’re an individual investor doing occasional research, where a frontier assistant on its own covers what you need.

	ChatGPT for Financial Services	Claude for Financial Services	Kepler (inside either, or its own app)
Where each number comes from	Written by the model from included and connected data	Written by the model from connected data	Pulled from source and computed by code
What a citation points to	The table or passage	The underlying data or document	The exact page, line or cell, checked before you see it
Your firm’s definitions	From what you connect, upload or template	From what you connect or upload	Built into your firm’s ontology, applied every time
Finding everything	Searches included data and connectors	Searches connected sources	Every document read in full and indexed before you ask
Same question twice	Can return a different number	Can return a different number	The same number, with the same sources
As filed vs as restated	Depends on the source	Depends on the source	Either, on any date
Your templates	Admins publish firm Excel, Word and PowerPoint templates	Follows the workbook’s formula conventions and the deck’s slide masters	Saved once as playbooks: comps, models and notes in house format every time
Where it works	ChatGPT, Excel, Google Sheets	Claude apps, Excel, PowerPoint, Word, Outlook	App, Excel, PowerPoint, Word, ChatGPT, Claude, Slack, API

See how the full stack fits together on the platform page, or read what enterprise intelligence means for firms like yours.

Try Kepler at app.kepler.ai, or book a demo and we’ll run it inside the assistant your team already uses.

Frequently asked questions

Is Claude or ChatGPT better for financial analysis?

Claude leads the most-cited public benchmark: on Vals AI's Finance Agent leaderboard updated October 29, 2026, Claude Opus 5.5 ranks first at 67.85%, and OpenAI's highest-scoring model on it, GPT-6.1 Sol, scores 61.82%. ChatGPT for Financial Services leads on included data, with filings, transcripts and financial datasets in the plan without separate contracts. For numbers a firm acts on, the stronger choice is the Kepler platform, which runs inside both and computes every figure in code with a citation to the exact line.

What is the best LLM for financial analysis?

On Vals AI's Finance Agent v2.1, which tests entry-level analyst work on public company filings, the top three as of October 29, 2026 were Claude Opus 5.5 (67.85%), Google's Gemini 4 Argon (67.66%) and Meta's Wonder Full (67.58%). The best Financial Modeling score was 57.0%. For numbers a firm acts on, the model matters less than whether code computes the figures and cites each one to its source.

Which has the better Excel add-in, ChatGPT or Claude?

Both are generally available. ChatGPT for Excel, GA since May 5, 2026, builds and updates models, links answers to the cells it uses and asks before editing; it also works in Google Sheets. Claude's add-ins cover Excel, PowerPoint and Word, with Outlook in beta, and carry context from one app to the next. With Kepler, statements and comps arrive in Excel with every cell cited.

Do ChatGPT or Claude train on my firm's financial data?

Not by default on their business plans. OpenAI says business data in ChatGPT for Financial Services isn't used to train its models by default, and Anthropic says it doesn't train on Claude Team or Enterprise data by default. Ask either vendor for its audit reports and data retention settings, and confirm how permissions and information barriers are set up.

Can I use Kepler with ChatGPT or Claude?

Yes. Kepler runs inside Claude through a verified MCP connector and inside ChatGPT, so analysts keep the assistant they prefer. When a question needs firm data or numbers, the answer comes from your firm's ontology, with every figure computed in code and cited to the exact page, line or cell. Anthropic wrote a case study on how we built Kepler with Claude.