ChatGPT vs Claude vs Perplexity vs Copilot vs Gemini for Finance and Portfolio Questions (2026)

The fastest way to see the split is dimension by dimension. The standout rows are about your real account. ChatGPT (through Finances, US only) and Perplexity (through Portfolio) can read a linked brokerage; Claude can through a connector; Copilot and Gemini cannot. “Places trades” is No for all five. Walnut, an AI investing app that knows your holdings, is the only “Yes” on trades, and its connection also works inside Claude and ChatGPT. After that, the assistants diverge on live data, citations, reasoning depth, and math.

Capabilities and pricing change, and the “with browsing on” or “with code tool” qualifiers matter: ChatGPT and Claude only fetch live prices when their search tool is active, while Perplexity, Copilot, and Gemini ground answers in the web by default. Verify the current feature set on each provider's site.

Each assistant below is described the same way: what it is, its strengths for investing, its weaknesses, who it suits, and one honest limitation. They split by job rather than by an overall ranking, because the right one depends on the question you are asking.

For live, cited research the best of the five is Perplexity, and for close reading of a long filing it is Claude. “Research” covers several jobs, so here is the pick for each:

• Current facts and quotes with sources. Perplexity wins. Live web search and an inline citation on almost every claim are its default, and its Finance view surfaces current prices, fundamentals, and news for a ticker. It is the one to reach for when you need today's number and want to see where it came from.
• Reading and reasoning over long documents. Claude wins. Its long-context strength means it can take a full 10-K, an earnings transcript, or a prospectus and reason across the whole thing without losing the thread, and its tone flags uncertainty instead of bluffing.
• All-round help and reliable calculations. ChatGPT wins. It is the strongest generalist for learning, drafting, and brainstorming, and its code-interpreter sandbox runs real Python, so allocation and percentage math is computed rather than guessed when you ask it to use the tool.
• Spreadsheet modeling and a Microsoft workflow. Copilot wins. If you already model a portfolio in Excel and live in Windows and Edge, grounded answers plus spreadsheet help in one place is its edge.

The honest takeaway: many investors use more than one, gathering cited facts in Perplexity, reasoning over them in Claude or ChatGPT, and checking the math in a code tool or Excel. None of them, used alone, can do the one thing that makes the analysis about you rather than a generic example.

Math reliability comes down to whether the tool computes or guesses. Large language models predict text, so when they do arithmetic in prose they can get a percentage, a position size, or an allocation subtly wrong while sounding certain. The fix is making them run real calculations:

• ChatGPT is most reliable when you ask it to use its code (Python) tool, which executes actual arithmetic rather than predicting the answer.
• Copilot is reliable inside Excel, where the spreadsheet does the computing and Copilot helps build the formulas.
• Perplexity tends to pull numbers from retrieved sources, which is safer than freehand math but only as good as the source.
• Claude reasons carefully and flags uncertainty, but in pure prose it can still slip on arithmetic, so verify anything that drives a decision.
• Gemini can compute in a linked Google Sheet, which is safer than freehand math, but in plain prose it carries the same slip risk as any model.

The outside evidence says to take this seriously. Saturn's 2026 “Artificial Authority” benchmark, as reported by Tom's Guide and the Financial Times, asked 17 models 121 personal-finance questions five times each and found 43% average accuracy, with free models wrong 63% of the time and paid models 49%. Errors included bad calculations and invented rules.

For any number that touches real money, the rule across all five is the same: have the tool compute it, then check it. A confident wrong figure is the most dangerous output an AI assistant can give an investor. In Walnut, the numbers about your holdings (position values, returns versus the S&P 500) are calculated from your broker's positions and live prices; the AI explains them rather than producing them. That does not make it immune to error, so check anything you act on.

Two can read them now. In the US, Finances in ChatGPT links accounts through Plaid and shows investments, current value and position size. Perplexity's Portfolio feature, launched in March 2026, links brokerages through Plaid too. Neither places trades. Claude reads a brokerage only through a connector you add, and Microsoft Copilot and Google Gemini have no brokerage connection of their own. The mechanism is simple: without a linked account or a connector, a chatbot only receives the text in the chat window, so unless your holdings are in that text, it has no way to know them. Without that link, three things follow:

• The analysis is generic. Ask any of them “is my portfolio too concentrated?” and the answer is about a made-up example, because the tool has no idea what you actually hold.
• You become the data pipe. The only way to make the conversation specific is to describe or paste your positions by hand, which is tedious, error-prone, and instantly out of date.
• They cannot act. None can place a trade or rebalance, ChatGPT Finances and Perplexity Portfolio included. They can describe a plan, not execute one.

The professional finance offerings do not change this for an individual. Claude for Financial Services and the finance-data integrations from providers such as LSEG, FactSet and Morningstar are built for firms; none of them reads your personal brokerage account. Reading a retail account takes ChatGPT Finances or Perplexity Portfolio (read-only), or Walnut, which reads and places trades you approve at Public, plus eight other brokers including E*Trade, tastytrade and Webull, in its own assistant or through its connector inside Claude and ChatGPT. For what ChatGPT now reads and what it still cannot do, see what ChatGPT Finances can and can't do.

“What does today's news mean for my holdings?” needs two things in one answer: your actual positions and a live search. Here is how each assistant handles it without you pasting anything:

• ChatGPT: only through Finances, in the US, read-only. Elsewhere you paste your holdings.
• Perplexity: through Portfolio, if you link a brokerage; its live search is the strongest of the five.
• Claude: only with a brokerage connector added; otherwise you paste.
• Copilot and Gemini: you paste your holdings each time.
• Walnut: its assistant reads your connected holdings and searches the web in the same answer, across every broker you connect.

Every comparison above ends at the same wall. The five assistants differ on live data, citations, reasoning, and math, and they are identical on the row that matters most for a real investor: none can act on what you actually own. ChatGPT and Perplexity can read a linked account; none of the five can place a trade, hold target weights or buy toward them.

Walnut, the site you are reading, is an investing chatbot that knows your holdings. It has its own AI assistant, and the same broker connection works inside Claude and ChatGPT, so you can ask about your actual positions wherever you already chat. Then it does the part the general assistants cannot: turns a thesis into a basket with target weights, runs recurring buys toward those weights with cash already at your broker, gives a keep, trim or exit read on each holding against the S&P 500, and places the trades you approve at Public, plus eight other brokers including E*Trade, tastytrade and Webull (US and Canada). Trades happen in the Walnut app, never through the connector, and every one needs your explicit approval.

To put it plainly: Copilot and Gemini cannot see your holdings, Claude sees them through a connector, ChatGPT and Perplexity can read them, and Walnut can read them and act on them with your approval.

To be upfront, since this is our site: Walnut does not try to out-reason Claude on a 10-K or out-cite Perplexity on today's news. Its assistant is about the portfolio you hold. Its dashboard frames each holding's return against the S&P 500 and classifies it as outperforming, in line, or lagging, alongside momentum and concentration reads. Because broker feeds rarely pass cost basis, Walnut frames returns as window returns rather than realized profit and loss, and says so. Walnut is not an investment adviser. If you want the mechanics, see how to connect your brokerage to an AI assistant.

Perplexity is the pick for cited research, Claude for long documents, ChatGPT as the all-rounder, and Walnut for anything about the portfolio you hold. The strongest setup usually uses more than one.

• You want today's facts with sources. Perplexity, for live quotes, recent news, and a citation on every claim you can click through to verify.
• You want to read a long filing or transcript closely. Claude, for long-context reasoning that holds the whole document at once.
• You want an all-rounder and trustworthy math. ChatGPT, for learning, drafting, and calculations run through its code tool.
• You live in Excel and Microsoft 365. Copilot, for grounded answers plus spreadsheet help where you already work.
• You live in Google Search and Workspace. Gemini, for search-grounded answers plus Sheets and Docs integration.
• You want the conversation to be about your actual holdings. ChatGPT Finances (US) or Perplexity Portfolio, if you only need to read them. To act on them, or to use Claude, none of the five alone: Walnut's assistant knows your holdings across brokers, its connection works inside Claude and ChatGPT, and it places any trade you approve.

Walnut owns the questions about your own holdings. For these other investing jobs, one of the general assistants (or a different product) fits better:

• You only want to learn or brainstorm with no money on the line. ChatGPT or Claude alone is enough; there is nothing to connect yet.
• You want fast, cited research on a ticker or the market. Perplexity is purpose-built for that, with sources on every claim.
• You want to model scenarios in a spreadsheet. Copilot inside Excel is the natural fit.
• You do not want to connect a brokerage at all. Walnut sits on top of a real account, so it needs one; a general assistant you paste holdings into would suit better.
• You want a one-shot risk score and recommendation report. A dedicated analyzer like PortfolioPilot delivers that format; see the best AI portfolio analyzers roundup.

From a connected account you can dig into a specific stock, an ETF you hold, or a theme you want exposure to.

For investing and portfolio questions, the pick among ChatGPT, Claude, Perplexity, Microsoft Copilot, and Google Gemini follows the job. Perplexity owns current, cited research. Claude owns deep reasoning over long filings. ChatGPT is the best all-rounder and the most reliable for math through its code tool. Copilot fits the Excel and Microsoft 365 crowd, and Gemini fits Google Search and Workspace users. But they share a hard limit: none can act on your real portfolio. ChatGPT (Finances, US) and Perplexity (Portfolio) can read a linked account; the rest work on a hypothetical until you connect one. For questions about the portfolio you hold, Walnut is the AI investing app that knows your holdings, inside its own app or through Claude and ChatGPT, and places the trades you approve at Public, plus eight other brokers including E*Trade, tastytrade and Webull (US and Canada). None of these tools, Walnut included, is an investment adviser.