Binance opened Binance AI Pro to every user on Oct. 191. The tool takes a trading idea in plain words and turns it into a strategy that runs on the exchange. A free Standard plan carries 250,000 credits a month, and a Premium plan at 19.99 USDC carries 10 million. The product had run as a paid beta since March 256. Jeff Li, Binance's VP of Product, called the launch a matter of access2. The CryptoLately.com Desk read Binance's announcement, its plan table and the beta's terms. It also read a study of what a language model does when told to make money.
Oct. 19, 2026: Premium's 10 million credits are 40 times the free plan's 250,000
Monthly credits in each Binance AI Pro plan, per Binance's plan table and its March beta terms
- The free plan: 250,000
- The paid plans
Sources: Binance [1]; Binance, via Zawya [6].
Read the chart as a table
| Item | Value |
|---|---|
| Standard, free | 250,000 |
| Beta, $9.99 | 5,000,000 |
| Premium, 19.99 USDC | 10,000,000 |
What Shipped on Oct. 19
A user types an idea: the assets, the conditions and the risk limits1. Binance AI Pro fills the gaps and draws the strategy as a workflow. On desktop the chat sits beside a Strategy Canvas, where a click on any step opens it for changes2. In the app, the building happens in chat. Paper trading then runs the strategy on hypothetical funds.
Both plans can send a strategy to live trading1. The free plan runs on DeepSeek. Premium adds frontier models, which Binance names as ChatGPT Sol, ChatGPT Astra and Claude Opus, where available, plus credit top-ups. That plan carries 10 million credits a month, 40 times the free allowance. Top-up credits expire after 12 months2.
Two limits sit in the release. Each user may hold ten strategies in total across paper trading, live trading and signals2. Backtesting against past prices is absent from the first release, so paper trading and an automated code test stand in for it. Spot and USDS-M futures are the markets covered, subject to regional restrictions. Simon Yoon of TokenPost noted that AI Pro runs on Binance.com, which stays closed to U.S. residents12. Binance.US sits outside the rollout.
Seven Months of Beta
The beta opened on March 25 at $9.99 a month, against a regular charge of $29.996. Subscribers got 5 million credits a month and a seven-day trial. That version was built on the OpenClaw open-source ecosystem and ran on ChatGPT, Claude, Qwen, MiniMax and Kimi. Binance Agent OS, a developer platform that lets outside agents trade through sub-accounts, followed on Aug. 204. On Oct. 5, co-CEO Richard Teng and Li unveiled Binance Intelligence in a livestream7. The brand bundles Binance AI, AI Pro and Agent OS. Binance AI Pro went free on Oct. 19, 208 days after the March 25 beta.
Oct. 19, 2026: AI Pro reached every user 208 days after its March 25 beta
Dated steps in Binance's AI stack, from the CFTC's bot advisory to the free launch
Read the chart as a table
| Date | Event |
|---|---|
| Jan. 25, 2024 | CFTC bot advisory. The CFTC's customer advisory on AI trading bots |
| March 25, 2026 | AI Pro beta, $9.99. 5 million credits, limited users |
| Aug. 20, 2026 | Agent OS. Outside agents trade via sub-accounts |
| Oct. 5, 2026 | Binance Intelligence. Teng and Li livestream the brand |
| Oct. 19, 2026 | AI Pro for all. Free Standard plan, 250,000 credits |
Finance Magnates saw a gap in the Oct. 5 plan. “Binance has not explained whether the forthcoming release will replace the existing beta, extend it to more users or introduce a different pricing model,” Tanya Chepkova, a news editor at the outlet, wrote on Oct. 55. On Oct. 19 the answer was all three. The beta stays alive, every user gets in, and the list price fell from $29.99 to zero1. Beta users keep their subscriptions and may move their chat history and strategies across.
Teng framed the stack as a knowledge project. “Finance has always been built for the people who already understand it,” he said in the Oct. 5 release7. “Blockchain is giving everyone access to markets; AI gives everyone the knowledge to navigate them.”
Rival exchanges built the same doorway this year. Kraken shipped an open-source command-line tool with a built-in MCP server in March, Jagmeet Singh reported in TechCrunch4. Coinbase followed in June with Coinbase for Agents, and OKX added an MCP toolkit. Robinhood, the brokerage Vlad Tenev co-founded and runs, put AI trading agents inside its app in late September, Unchained added315.
Where the Money Sits
Every live strategy runs inside its own sub-account, an AI Strategy Account2. Only the user can fund it, by manual transfer. The agent has zero power to move funds, and the sub-account sends money only to the user's other accounts. A strategy trades only after the user approves it, and an edit creates a new version that needs approval again1. The user can pause, adjust or revoke it at any time.
The code check is narrower than it sounds. Binance's check confirms the logic runs free of errors, Unchained reported3. In the words of Binance's blog post, it "does not assess profitability, quality, or likely performance" of a strategy. Maisie Morrison of Blockonomi added the fee point13. The word "free" covers the software, while users supply the trading funds and pay Binance's trading fees.
Li described the same sub-account design to TechCrunch in August, for Agent OS. “Instead of total freedom, we put the power in users' hands to give them the granular access control of what they can do through the agent,” he said4. Asked whether Binance can see why an agent trades, he said the reasoning happens outside its systems. “We really cannot see the reasoning of what the user's action is,” he said. TechCrunch reported that the amount moved into the sub-account serves as the only limit on what an agent can lose.
AI Pro differs on that point. Binance AI Pro pairs a language model with a finance-specific agent harness Binance built2. The harness adds guardrails and a workflow generator that writes the strategy code. Read against Li's August answer, that puts the model's reasoning inside Binance's own harness this time.
The Case For and the Case Against
Li makes the case for. “Binance AI Pro is about access: professional-style strategy building, no coding required, with the user in control at every step,” he said in the launch release2. “Users describe the idea, review the workflow, test it with paper trading, and approve deployment themselves.” The pitch rests on the approve button: the agent writes, the person decides.
The strongest case against comes from a laboratory. Seungpil Lee, Donghyeon Shin, Yunjeong Lee and Sundong Kim work at the Gwangju Institute of Science and Technology9. The team ran four language models through a slot machine with a negative expected value. “In slot machine experiments, we identified cognitive features such as illusion of control and loss chasing, observing that greater autonomy in betting parameters substantially amplified irrational behavior and bankruptcy rates,” they wrote. Jose Antonio Lanz of Decrypt reported the figures10. Gemini 2.5 Flash went broke in 48 percent of sessions once it set its own bets. GPT-4.1-mini went broke in 6.3 percent. Prompts that said "maximize rewards" pushed the models toward all-in bets.
2025 study: Gemini 2.5 Flash went broke 48 percent of the time when it set its own bets
Share of slot-machine sessions ending in bankruptcy, per Decrypt's report on the Gwangju study
- Highest bankruptcy rate: 48.0%
- Lowest reported rate: 6.3%
Source: Decrypt [10].
Read the chart as a table
| Item | Value |
|---|---|
| Gemini 2.5 Flash | 48.0% |
| GPT-4.1-mini | 6.3% |
The study covered 12,800 gambling sessions, each starting from $10010. Its models were GPT-4o-mini, GPT-4.1-mini, Gemini 2.5 Flash and Claude 3.5 Haiku, smaller cousins of the models on Binance's Premium list.
Practitioners say the same in plainer words. Nodari Kolmakhidze, chief financial officer of Cindicator, which built Stoic.AI, told Cointelegraph: “The biggest misconception is that AI bot is like money printer… It's not like that”8. Brett Singer, sales and research lead at Glassnode, said most bots "did not beat the market" and that today's AI works best as “an associate or an intern that can work 24 hours a day.”
Regulators got there first. The CFTC's customer-education office published an advisory on Jan. 25, 2024, titled "AI Won't Turn Trading Bots into Money Machines"11. “When it comes to AI, this advisory is telling investors, 'Be wary of the hype,'” said Melanie Devoe, the office's director. The advisory's case study was a South African operator who promised 10 percent a month from a bot. He took more than $1.7 billion in bitcoin from at least 23,000 people. Binance promises the opposite: its disclaimer says AI outputs may carry errors, biases and outdated information1. Users bear every loss, it adds.
Li's case holds if an approve button and a funded sub-account cap a trader's losses at what they chose to risk. Lee's team makes the other case. A model writing the strategy may carry the biases they measured, and a trader may approve a workflow unread. Each reader decides which trader they are.
By the numbers
- 250,000: credits a month on the free Standard plan, against 10,000,000 on Premium for 19.99 USDC1.
- Ten: strategies a user may hold across paper trading, live trading and signals2.
- $9.99: the beta's monthly price from March 25, against a $29.99 regular charge6.
- 5,000,000: credits a month in that beta, against 250,000 on the free plan today6.
- 208 days: from the March 25 beta to the Oct. 19 release.
- 48 percent: Gemini 2.5 Flash's bankruptcy rate when it set its own bets, against 6.3 percent for GPT-4.1-mini10.
- 12,800: gambling sessions in the Gwangju study10.
- $1.7 billion: bitcoin taken from at least 23,000 people in the CFTC advisory's example11.
- 280,000: daily calls Binance says Agent OS already handles, per Cointribune14.
What to watch
Backtesting is the first signal, since Binance left it out of the first release2. Its arrival would show the product growing. The regional list is the second. Finance Magnates and TokenPost both flagged that Binance has yet to publish which countries get live trading512. A third is the paper-trading gap Cointribune named, because a simulation skips slippage and execution speed14. Readers who want the AI side of this story can read sister publication AI Lately at ailately.com15. AI Lately covers the models, the agents and the labs behind them. Its Oct. 15 column asked who keeps the gains when bots trade on exchanges.
Sources
- Binance, "Introducing Binance AI Pro," Oct. 19, 2026, https://www.binance.com/en/support/announcement/detail/c50ffc85804f45f08d0d7727ea508157
- Binance, "Binance Launches Binance AI Pro, Bringing Conversational AI Strategy Building to Users," via PR Newswire, Oct. 19, 2026, https://www.prnewswire.co.uk/news-releases/binance-launches-binance-ai-pro-bringing-conversational-ai-strategy-building-to-users-302912585.html
- Unchained, edited by Jeff Albus, "Binance Lets All Users Run AI-Built Trading Strategies Following Successful Beta," Oct. 19, 2026, https://unchainedcrypto.com/binance-lets-all-users-run-ai-built-trading-strategies-following-successful-beta/
- Jagmeet Singh, "Binance now lets AI agents trade, but keeping them in check is largely up to users," TechCrunch, Aug. 20, 2026, https://techcrunch.com/2026/08/20/binance-now-lets-ai-agents-trade-but-keeping-them-in-check-is-largely-up-to-users/
- Tanya Chepkova, "Binance Launches Retail AI Assistant, Repackages Existing Trading Tools," Finance Magnates, Oct. 5, 2026, https://www.financemagnates.com/cryptocurrency/exchange/binance-launches-retail-ai-assistant-repackages-existing-trading-tools/
- Binance, "Binance Beta launches Binance Ai Pro, bringing AI Agentic trading to users," via Zawya, March 30, 2026, https://www.zawya.com/en/press-release/companies-news/binance-beta-launches-binance-ai-pro-bringing-ai-agentic-trading-to-users-351690
- Binance, "Binance Launches Binance Intelligence to Close the Financial Knowledge Gap," via PR Newswire, Oct. 5, 2026, https://www.prnewswire.co.uk/news-releases/binance-launches-binance-intelligence-to-close-the-financial-knowledge-gap-302898440.html
- Savannah Fortis, "AI trading bots surge in popularity, but experts warn they're not 'money printers'," Cointelegraph, Nov. 7, 2025, https://cointelegraph.com/news/ai-trading-bots-surge-experts-warn-not-money-printers
- Seungpil Lee, Donghyeon Shin, Yunjeong Lee and Sundong Kim, "Can Large Language Models Develop Gambling Addiction?," arXiv 2509.22818, revised Dec. 19, 2025, https://arxiv.org/abs/2509.22818
- Jose Antonio Lanz, "Your AI Trading Bot Might Have a Gambling Problem," Decrypt, Nov. 6, 2025, https://decrypt.co/347623/your-ai-trading-bot-gambling-problem
- CFTC Office of Customer Education and Outreach, "CFTC Customer Advisory Cautions the Public to Beware of Artificial Intelligence Scams," Release 8854-24, Jan. 25, 2024, https://www.cftc.gov/PressRoom/PressReleases/8854-24
- Simon Yoon, "Binance AI Pro Requires Manually Funded Accounts for Live Trading," TokenPost, Oct. 19, 2026, https://www.tokenpost.com/news/technology/34842
- Maisie Morrison, "Binance Launches Free AI Pro Plan for Automated Crypto Trading," Blockonomi, Oct. 20, 2026, https://blockonomi.com/binance-launches-free-ai-pro-plan-for-automated-crypto-trading
- Luc Jose A., "Binance Rolls Out AI Pro Trading Agent In October 2026," Cointribune, Oct. 6, 2026, https://www.cointribune.com/en/binance-rolls-out-ai-pro-trading-agent-in-october-2026/
- Ryan Elliott Dennis, "Crypto Exchanges Let the Bots Keep the Gains," AI Lately, Oct. 15, 2026, https://ailately.com/articles/crypto-exchanges-bots-keep-the-gains