Binance AI Pro turns a chat into a trading strategy. The launch notes say what it can't do yet

On October 19, 2026, Binance launched Binance AI Pro, which it calls a finance-focused AI agent that turns trading ideas into tested, automated strategies through conversation. We read the company's press release in full. This is a read of the announcement, not a hands-on test, and it comes from the vendor, so treat every capability as a claim.

What it is

Users describe assets, conditions, parameters and risk limits in plain language. The agent fills gaps, suggests improvements and shows the result as a visual workflow, which Binance calls the Strategy Canvas. On desktop the canvas sits next to the chat and each step can be inspected or edited. On mobile, everything happens through chat. Binance says the agent writes the code and the user decides whether, when and what to run.

Under the hood, Binance says it pairs a large language model foundation with a finance-specific agent harness it built, with guardrails and a workflow generator that converts plain-language ideas into executable strategy code. The release does not name which model powers the free tier.

The safety design, as described

The release lays out a staged path. Strategies can be tested with paper trading, which uses virtual funds. Going live needs a separate review-and-approve step and runs in a dedicated AI Strategy Account, a virtual sub-account isolated from the main account.

Only the user can fund that sub-account, by manual transfer. Binance says the agent cannot move funds by itself, strategy sub-accounts cannot withdraw to external addresses, and strategies can be paused, adjusted or revoked at any time. Binance links this design to a policy paper it published earlier this month called Know Your Agent, which argues that agents should analyze and recommend while consequential actions need explicit human approval.

What is missing

The notes to editors carry the limits. Backtesting is not available in the initial release. Instead there is automated code testing and paper trading. That matters because a backtest asks how a rule would have behaved on past data, while paper trading only shows what happens from today forward, which takes time to say anything.

Other limits listed: up to 10 strategies per user across paper trading, live trading and signals; support for Spot and USDS-M Futures strategies only, subject to regional restrictions; more products to be added progressively.

The legal disclaimer is equally direct. Binance says use is at the user's own risk, that outputs may contain errors, biases or outdated information, that outputs are not advice, and that users are responsible for their prompts. It also says prompts may be used for training purposes.

Pricing

Everyone starts on the free Standard Plan with 250,000 monthly credits. The Premium Plan costs 19.99 USDC per month, raises the allowance to 10 million monthly credits, allows top-ups and adds frontier models, where available. Binance names ChatGPT Sol, ChatGPT Astra and Claude Opus. Premium top-up credits do not refresh, expire after 12 months and are used only after the monthly allowance. The release does not say how many credits a typical strategy build costs, so the real price of use is unknown.

How it compares with what we have covered

Interactive Brokers and Schwab have each described how outside AI agents may connect to accounts, a question we covered in our Schwab agent piece. Binance has taken a different route: its own agent, inside its own walls, with a sandboxed account. The common thread is the same, a human approval step between the model and the money.

The test we would run next is simple and costs nothing: give the agent one plain-language rule, run it in paper trading for a month, and compare what it did with what the rule said. No real funds, no results published until the month is over.