Grok searches X as posts come in, reads a chart and writes code. Sending an order is a different job: it takes a broker connector, the right settings and an account you can afford to lose. Here is what works in November 2026, which brokers name Grok, and what happened when Grok traded real money in nof1’s Alpha Arena.
November 16, 2026
19
min to read
The short answer
Not by itself. Grok can search X, read a chart and run code, but it has no brokerage account. Orders go through a broker connector you add, such as Robinhood’s, Interactive Brokers’ or Coinbase’s, or code you write with xAI’s API. In nof1’s 2025 contests, Grok 4 lost 45% in the first season; Grok 4.20 won the second.
Grok is the AI assistant made by xAI, which SpaceX acquired on February 2, 2026, according to SpaceX’s quarterly report; xAI’s developer docs now sign as SpaceXSI. As of November 15, 2026, its newest model is Grok 4.7, released on xAI’s API on September 21, 2026. It reads text and images, can hold 500,000 tokens (the word pieces models read) in one conversation, and its training data runs to May 2026.
You meet Grok in three places, and each one gives a trader something different:
| Where | What it adds for a trader | Where it stops |
|---|---|---|
| The Grok apps (grok.com, iPhone, Android) | Chat, voice, uploads of spreadsheets and PDFs, and connectors to outside tools; the catalog includes an S&P Global connector for financial data and company information | No brokerage account; a broker is a connector you add yourself |
| Grok Bot, launched August 11, 2026 | An agent with its own computer in the cloud: it uses apps and websites, runs routines on a schedule and asks for your OK when its checks or your rules flag an action | Routines run at least five minutes apart, and a website it reads is a website it can misread |
| The xAI API | Grok 4.7 with web search, X search, code execution, function calling and connections to remote MCP servers, for programs you write | Your program places the orders, bugs included |
For trading, four jobs stand out:
None of this needs a broker. Ready-made prompts for these jobs are in our library of trading prompts, and they work in Grok as written.
Without search, Grok knows nothing after its training data ends. xAI’s model docs say it plainly: “Grok has no knowledge of current events or data beyond what was present in its training data,” and real-time data needs the Web Search or X Search tools. With them on, Grok can work with what was posted this morning.
The X Search tool, as xAI documents it for developers, gives the model four moves: keyword search with X’s advanced operators, search by meaning, a search for accounts, and fetching a post with its parents and replies. You can narrow it to up to 20 handles, or exclude up to 20, and limit it to a date range. Looking at images and videos inside posts is off by default. xAI notes that your handle and date limits apply to every keyword and semantic search the model writes, which makes them the safest way to keep it on topic.
That makes Grok good at the questions X answers well:
And weak at the questions X answers badly. A post is a claim, not a fact, and a feed of posts is not a price feed. Promoters post too: the CFTC, in its advisory on AI trading bots, tells traders to “be wary of the hype around SI especially when promoted by social media influencers and strangers you meet online” (the CFTC writes SI, for super intelligence, where most people would say AI). Text can also be written to steer an AI that reads it, which is called prompt injection. xAI designed Grok Bot’s approval checks so that text from somewhere else, “such as an email or a web page, doesn’t count as your request,” and you should hold your own reading to the same rule.
Search X for posts about [TICKER] from [START DATE] to [END DATE]. Sort what you find into three groups: news from the company or a regulator (link the original post or filing), reporting by news outlets, and opinions. For each group, give the three most shared posts with the handle, the time and the link. Flag any account that promotes a token, a paid group or a signal service. Then tell me which claims a primary source confirms and which are only repeated. Don’t tell me whether to buy or sell, and don’t predict the price.
Not out of the box. Grok has no brokerage account and no orders resting at any venue. It can reach an account in three ways, and you set up each one yourself: a broker connector in the Grok apps, code you write on xAI’s API, both covered below, and Grok Bot on a broker’s website, which carries the most caveats.
xAI describes Grok Bot as “an agent with access to a real computer in the cloud, so it can do all the things you can.” In principle, that includes logging in to your broker and pressing Buy. None of the use cases xAI lists for Grok Bot involve trading. What its docs do say is to set explicit boundaries for actions such as “Purchases and financial transfers,” to “Start with read-only tasks and draft outputs,” and that Auto Review, the check that decides when the Bot stops to ask you, “is model-based and should complement, not replace, least privilege and explicit approval boundaries.” Grok Bot also picks its own model for each task; xAI’s FAQ says it can be an xAI model or a third-party one, so the agent on your broker’s site may not be Grok at all.
The brokers are blunter. Interactive Brokers warns on its AI integration page: “If you link an AI provider, including browser-based and similar agents, you are providing those AI agents with full control of your account, without any controls or limits.” A browser agent sees the same buttons you do, with none of the limits a broker builds into its own agent tools.
A Grok Bot routine runs a task on a schedule or when an event arrives. xAI’s routine docs say schedules “must be at least five minutes apart,” and its approvals guide adds that an approval raised by background work expires after 30 minutes if nobody answers. That is fine for a morning summary of your watchlist. It can’t protect a position: a market can gap through your level between two runs, and a stop that exists only in a routine’s instructions is not an order. Put stops and limits at the broker or exchange, where they work with every chat closed.
Robinhood, Interactive Brokers and Coinbase all name Grok on their own pages, as of November 15, 2026. Each one connects through MCP, the Model Context Protocol, an open standard that lets an AI app use another company’s tools, such as “get positions” or “place order.” In the Grok apps, a server that isn’t in the catalog goes in as a custom connector: you paste its URL and sign in on the broker’s own page.
| Broker | How you add it in Grok | What the connection can do | The built-in limit |
|---|---|---|---|
| Robinhood | Start a chat, select +, then Add connector, Custom, and paste https://agent.robinhood.com/mcp/trading | Reads all your Robinhood accounts, account numbers included, with positions and order history; trades stocks, options and crypto in a separate MCP account | It trades only with the money you deposit in that account. Trade approvals are off by default for MCP accounts |
| Interactive Brokers | Search for “Interactive Brokers” in Grok’s connector directory and sign in on IBKR’s login screen | Research, portfolio analysis, monitoring and trade instructions, which you review and submit yourself | Autonomous Trading, where the AI submits orders without asking each time, works only inside a funded AI Partition |
| Coinbase | Use the Connect button on Coinbase’s docs, or open Plugins, search for Coinbase, then Add and Connect; in Grok Bot, use the Marketplace | Spot crypto on more than 900 pairs, US futures and eligible US stocks | You pick which portfolios the agent can reach when you sign in |
The details are where the risk sits:
The menus differ from page to page: Robinhood starts from the + button, while Coinbase and TradingView start from Plugins. Follow the broker’s current page rather than an old screenshot. For market data alone, TradingView’s MCP server also lists Grok, added under Plugins, New Connector, Custom; it serves quotes, screeners, news and alerts on Essential plans and up, and has no tool for placing orders. How the same brokers work with other assistants is in our guide to agentic trading, and Robinhood’s own agents are covered in Robinhood Agents and its alternatives.
A “Grok trading bot” usually means a program that asks Grok for a decision and sends orders to a broker’s API. Grok is the reasoning; your code is the trader. xAI’s API gives that program three kinds of tools, and each one belongs in a different place:
require_approval, the approval setting in OpenAI’s version of this API, is not supported, and suggest using allowed_tools to “restrict access to tools that only perform read-only operations.” Give a broker’s server to Grok this way only with its order tools left out.Decide the market, the entry, the exit, the stop and the size in words a computer could follow. Algorithmic trading for beginners walks through it.
Before Grok makes a single call, run the rule on history with costs included, as in how to backtest a trading strategy.
Maximum size, allowed symbols, a stop with every entry and a daily loss cap belong in the order function, not in the prompt.
Run it on paper, or in a separate account that holds only what you could lose. Our guide to paper trading crypto covers testnets and demo accounts.
xAI’s examples read the API key from an environment variable; keep broker keys the same way, never in a message to the model.
Save what Grok saw, what it decided and what your code did, so a bad trade can be traced.
Nof1, an AI research lab, ran Alpha Arena, public contests in which AI models traded real money with no human stepping in. It ran two seasons in 2025, and Grok had a very different result in each.
In nof1’s own write-up, published October 27, 2025, the lab says it gave six models $10,000 each to trade on Hyperliquid “with zero human intervention”: GPT-5, Gemini 2.5 Pro, Claude Sonnet 4.5, Grok 4, DeepSeek V3.1 and Qwen3-Max. They traded perpetual futures, contracts with no expiry date that allow leverage, on six coins: BTC, ETH, SOL, BNB, DOGE and XRP. Every model got the same prompt and the same numerical market data, no news, and a new decision every two to three minutes. The season began on the evening of October 17, 2025, New York time, and ended at 5:00 p.m. ET on November 3.
| Model | Account at the close | Change from $10,000 |
|---|---|---|
| Qwen3-Max | $12,202 | +22.0% |
| DeepSeek V3.1 | $10,453 | +4.5% |
| Claude Sonnet 4.5 | $6,890 | −31.1% |
| Grok 4 | $5,456 | −45.4% |
| Gemini 2.5 Pro | $4,376 | −56.2% |
| GPT-5 | $3,734 | −62.7% |
| Bitcoin bought and held (nof1’s benchmark) | $9,994 | −0.1% |
The figures are nof1’s account totals at the close, as its site served them and the Wayback Machine saved them on November 4, 2025. Grok 4 finished fourth of six, down 45.4% while nof1’s Bitcoin benchmark was flat. Nof1’s write-up also describes how each model traded. Grok 4, GPT-5 and Gemini 2.5 Pro went short “much more frequently than peers.” Grok 4 was typically the least active, held positions the longest in pre-launch runs, and, with DeepSeek, set the loosest stops and targets.
The second season started on November 19, 2025, according to a post by nof1’s founder reported by ForkLog. This time the models traded only US stocks, with $10,000 of real money per model in each of four competitions, and each competition had its own rules: New Baseline, with news and memory; Monk Mode, focused on preserving capital; Situational Awareness, where models could see their rivals’ rankings; and Max Leverage, which required high leverage on every trade. Two newcomers joined, Kimi 2 and an unnamed “Mystery Model.”
Alpha Arena’s site still carries nof1’s result: “The official competition has ended as of December 3rd, 2025 at 5:00 PM EST. Models are no longer running. Mystery Model was the winner with a 12.11% aggregate return in 2 weeks. In total across competitions, it made $4,844.” Nof1’s leaderboard names the winning model as Grok 4.20, and on November 15, 2026 it showed Grok 4.20 above its $10,000 start in all four competitions. The older Grok 4 was below its start in all four, and its $385.02 in Situational Awareness, down 96.15%, was the lowest of the leaderboard’s 32 entries. xAI released Grok 4.20 on its API on March 10, 2026, according to its release notes.
Nof1 itself warned before the first season ended that “early successes may be the result of luck,” and listed the flaws of its own test: “prompt bias, limited sample sizes / lack of statistical rigor, and shortness of evaluation period.” Two seasons of a few weeks each, on one harness, can’t rank models as traders; models from the same family finished with a 45% loss in one season and first place in the next. The lab’s home page still sums up its view of today’s models: they “are shockingly bad at navigating financial markets.” Nof1 now gives Alpha Arena its own site, alphaarena.ai; the figures here come from nof1’s pages for the two 2025 seasons.
Most of the ways Grok can hurt a trade come from the same place: it reads a lot, fast, and it acts on what it reads. These are the failure points, and what limits each one.
| Risk | How it shows up with Grok | What limits it |
|---|---|---|
| Posts taken as facts | A rumor on X summed up as news | Ask for the primary source, such as a filing or the company’s own account, before you act |
| Old knowledge | An answer from training data that ends in May 2026 | Keep search on, ask for the date of every number and check the price at your broker |
| Text written to steer it | A post or a page with instructions aimed at the AI | Research in a session that can’t reach your account; connect only servers you trust |
| Approvals that loosen over time | An “Always allow” on an action, which saves a rule so similar actions run on their own | Keep narrow Ask first rules for anything that sends, buys or deletes |
| Connections that see everything | A broker connector that reads all your accounts, or one shared by every Bot | A separate account or portfolio, funded with what you could lose |
| Stops that live in a chat | A plan to sell at a level that exists only in a conversation or a routine | A real stop order resting at the broker or exchange |
Two habits cover most of the table. Split reading from trading: do your X and web research in one chat, and give order tools, if you connect them at all, to another. And keep secrets out of the conversation. xAI’s security guide for Grok Bot puts it simply: “Do not send a password, one-time code, or API key in ordinary chat.” If a product promises that Grok will trade you to steady profits, read is AI trading legit? first; the CFTC’s advisory says this technology “can’t predict the future or sudden market changes.” The research on AI forecasts is in our guide to whether AI can predict the stock market.
Grok is a general assistant with an unusual window onto X. An AI trading agent is built around the trade instead: the market data, the account and the limits come with it. Ethy is one, the AI trading agent you ask before you trade, and it covers the gaps this page found:
It places the trade too. From chat, Ethy trades Hyperliquid perpetual futures, on crypto and on the stocks, indices, commodities and currencies listed on trade.xyz, and it runs spot swaps. Leverage magnifies losses as well as gains. A setting decides whether each chat order shows a confirmation card first; new users start with the card. Automations trade on their own once you activate them, or run in paper mode on live prices with no money at stake, and they can’t be triggered by news or sentiment, so a viral post can’t fire one. Stop loss and take profit orders sit on Hyperliquid and trigger even when Ethy is closed. Your balance lives in a wallet that belongs to you, and you can export its key at any time. You can’t connect Grok or any other assistant to Ethy; you use it in its own app. Ethy is software for research and analysis, and it doesn’t give investment advice.
Grok stays the better tool for what X does best: who said what, and when. Claude and ChatGPT answer the same question with their own connectors and limits. Our wider LLM trading comparison lines up Grok against Claude, ChatGPT and Gemini on the same tasks.
Not from xAI: its docs describe no trading product. A “Grok trading bot” is usually a program someone wrote that calls xAI’s API, or a third-party product using the name. If you build one, keep order placement in your own code with hard limits. If you buy one, test it on paper and walk away from anyone who promises steady returns.
Only through an account you connect. Coinbase’s MCP connects Grok and Grok Bot to spot crypto on more than 900 pairs, and Robinhood’s Trading MCP includes crypto tools for its separate MCP account. Crypto trades around the clock, so keep a real stop order at the exchange rather than a plan in a chat.
Yes, through Robinhood’s Trading MCP. In Grok, start a chat, select +, choose Add connector and Custom, and paste https://agent.robinhood.com/mcp/trading. The agent trades only in a separate MCP account you fund. Trade approvals are off by default for MCP accounts, so turn them on in Safety Controls if you want to place each order yourself.
Not by itself. xAI’s docs say Grok has no knowledge of current events beyond its training data unless search tools are on. X and web search return posts and pages, which can quote old or wrong prices. Ask for the source and time of every number, and check the price at your broker before you act.
One season. In nof1’s first season, on crypto in October and November 2025, Grok 4 finished fourth of six, down 45.4%. In Season 1.5, on US stocks, the “Mystery Model,” which nof1’s leaderboard names as Grok 4.20, won with a 12.11% aggregate return over two weeks. Nof1 warned that results this short can come down to luck.
No tool can do that reliably, and the evidence doesn’t point to Grok being an exception. Nof1, whose contests Grok has both lost and won, calls today’s language models shockingly bad at markets, and the CFTC says this technology can’t predict the future or sudden market changes. Use Grok to gather and check information, and test any rule before trading it.
Trading involves risk, including the loss of your money. Ethy AI is a research and trading tool and does not give investment advice or recommendations. Backtests are simulations on historical data and do not predict future results.
Deploy your AI trading agent in one click.
