← All tools · Guides

How to choose between QMT and PTrade: MiniQMT, fees, and a model API

A choosing guide. You have decided a program should send orders. QMT or PTrade?

How to choose: If you want your own Python libraries, outside data, or a model API, choose QMT and write the strategy with MiniQMT plus xtquant. If you do not want a computer left on, and the strategy uses only the platform's built-in data and libraries, choose PTrade. If you are still learning and only want a backtest, start with a hosted site such as JoinQuant or RiceQuant. If you want the data, the backtest, and the orders on machines you control, use an open-source framework such as vn.py. Neither terminal is sold as a download. The first step is an account at a broker that offers it. Thresholds and fees follow that broker's own disclosures. Paper-trade first, then a live account. This page compares QMT and PTrade only. It is not the six-platform overview, not the tool-order guide, and not the site FAQ.

Who it is for: Someone who has already decided a program should send orders on their own stock account, and is choosing between QMT and PTrade. What you need is where the code runs, whether you can install libraries and call a model API, what MiniQMT is, how to apply, whether there is a fee, and the programmatic-trading report in force from July 2025.

Who it is not for: Someone still comparing JoinQuant, RiceQuant, MyQuant, and BigQuant, or who wants the order of data, then a backtest, then paper trading. Thresholds and fees differ by broker, and this page does not publish a number for them. Opening a cloud server and connecting over SSH is a different article, and that walkthrough uses Ubuntu. QMT needs a Windows image. This page is not the six-platform overview and not the site FAQ.

The same guide in Chinese is how to choose between QMT and PTrade. The wider comparison of broker terminals and hosted sites is how to choose a quant trading platform. That comparison is not this article. An open-source backtesting framework on a machine you control is how to choose a backtesting framework. After a backtest passes, the two-week ledger is paper trading. A lightweight Ubuntu desk for crawlers and cron is the cloud quant desk. That desk does not run QMT.

This site has no commercial relationship with ThinkTrader, Hundsun, or any broker, and it does not offer account-opening links. Facts below follow the Chinese guide. They were checked on 21 Oct 2026 against the ThinkTrader knowledge base, the Shanghai Stock Exchange texts, and a third-party compilation of the PTrade docs. Opening conditions and fees follow the broker. This is not investment advice. Programmatic trading can lose money. A backtest is not a future return.

01 · One table

02 · QMT, MiniQMT, and the research terminal

The ThinkTrader beginner guide describes two products. Broker QMT is for trading, and you apply through a broker. The research terminal is for strategy research, and you register and buy it on ThinkTrader's research site. An individual who wants a program to send orders on their own stock account uses broker QMT.

Checking "minimal mode" at login enters MiniQMT. It is not mainly a charting app. It gives xtquant a connection. xtquant is ThinkTrader's Python library on top of MiniQMT. It has two modules, market data (xtdata) and trading (xttrader): place an order, cancel, query assets, and query positions. The docs say you must start the MiniQMT client before a program that uses xtquant. You install xtquant from a zip on the ThinkTrader knowledge base, not with pip. It supports 64-bit Python 3.6 through 3.12. The May 2025 release notes also list Python 3.13.

Leave minimal mode off and you get the full QMT client, often called big QMT. It embeds Python 3.6 and runs the strategy inside the client. NumPy and Pandas are included. Every strategy runs on the same thread. Multithreading and multiprocessing are not available.

People writing Python strategies mostly use MiniQMT plus xtquant. The strategy is an ordinary Python program. You choose the version, the libraries, and the debugger.

03 · PTrade's hosted mode: what you gain and what you give up

PTrade is Hundsun's quant trading terminal. Hundsun's site does not publish a manual. The points below come from a third-party compilation, the PTrade API docs. The version your broker runs controls:

• The strategy is hosted in the broker's machine room. It keeps running when your computer is off. You do not have to worry about power, the network, or the computer sleeping.
• Third-party libraries are a fixed list that changes with the platform version. The compilation has recorded arch and cvxopt being removed and PuLP being added. You cannot install other libraries.
• The platform has a Python 3.5 environment and a Python 3.11 environment. Some functions return a different shape in the two environments. Which one you get depends on the version the broker deployed.
• The style is close to JoinQuant, with functions such as initialize and handle_data. Backtests support daily and minute bars. Live trading also supports tick.

The cost: the code lives in the broker's machine room, and the data is what the platform provides. We did not find a public note we could check on whether that room can reach the public internet. Follow the broker's own disclosures.

04 · How to apply

The two flows are roughly the same:

1. Open an account at a broker that offers QMT or PTrade. If you already have an account, ask that broker first.
2. Apply to the broker. Sign the supplemental agreement on programmatic trading and fill in the report, as in section 07.
3. Download the client from the channel the broker gives you, and log in with the account the broker issues.

Thresholds: whether there is an asset threshold, and how high it is, differs by broker and sometimes by branch. Follow the broker. The ThinkTrader community keeps a list of which brokers support QMT, in high, medium, and low tiers (edited April 2025). Use it to see who offers the terminal, then ask for the actual conditions. Claims on the web of the form "this broker opens it at this asset level" were not checked broker by broker, and they are not copied here.

No similar public list was found for PTrade. Follow the broker there too.

05 · Fees

We did not find a public broker page that states a fee schedule for QMT or PTrade. Before you apply, ask three things: whether the software itself costs money; whether there is a monthly or yearly fee; and whether opening it is tied to the commission rate, assets, or volume. Follow the broker's own disclosures.

06 · A model API, or an AI strategy: which one

If the strategy should call an API such as DeepSeek or Qwen, or run a research agent such as TradingAgents-CN, the question is whether the code can reach the public internet and whether you can install the libraries it needs. The Docker deploy note is Chinese only.

• MiniQMT plus xtquant: the strategy is an ordinary Python program on your computer. Calling a model API, reading a local database, and using a machine-learning library such as Qlib all happen in your environment. The order goes through xtquant. A worked Qlib run is the Qlib walkthrough. Wiring market data and a model API is the Chinese note connecting a data source and an LLM API.
• Big QMT's built-in Python: the ThinkTrader docs say the broker can turn on a Python library whitelist. Installing a library outside that list raises an error, and you have to ask the broker for permission. Posts on the ThinkTrader community also say that many brokers have recently turned on a restriction against outside connections for the built-in Python. We did not find a formal notice from a broker or from ThinkTrader. Ask before you rely on it.
• PTrade: libraries are the built-in list only. Whether the room can reach the public internet follows the broker. If you want a model in the strategy, generate the signal on your own computer first, then see which way the broker allows that result into the strategy.

In short: if you want AI, prefer MiniQMT plus xtquant.

07 · The programmatic-trading report

The CSRC published the Securities Market Programmatic Trading Management Provisions (Trial) on 15 May 2024. The Shanghai Stock Exchange Programmatic Trading Management Implementation Rules took effect on 7 July 2025. Three points matter most for an individual:

• Report first, then trade. Under the SSE notice on reporting programmatic stock trading, an investor who conducts programmatic stock trading on the SSE must report to the broker that accepts the orders before doing so for the first time. The broker checks the report and files it with the exchange within five trading days. An investor who has not reported must not conduct programmatic trading. The report includes the strategy type, the highest order rate, and the highest number of orders in a day.
• High-frequency trading has extra requirements. In the SSE press Q&A, an account whose combined orders and cancels reach 300 or more per second at the peak, or 20000 or more across the day, is treated as high-frequency trading. It must also report the server location, a system test report, and a fault plan.
• An ordinary individual usually does not hit the abnormal-trading indicators. The same Q&A says that, during the trial, the accounts that triggered the indicators were mainly quant private funds and broker proprietary desks. Small individual investors basically did not.

The points above are the SSE rules. Trading on the Shenzhen Stock Exchange follows the SZSE rules. The broker tells you which set applies.

08 · The computer has to stay on

A QMT program runs on your own computer. During the session both the computer and MiniQMT have to stay on. The ThinkTrader install notes are written for Windows. Do not put the install directory on the C drive, or you may have to run as administrator. Two common setups follow: a Windows computer at home that stays on, or a cloud server with a Windows image.

The cloud-server note on this site, opening a server and connecting over SSH (Chinese), uses Ubuntu. It fits data collection, backtests, and a research agent. The English edition of that desk, which is a different article, is the cloud quant desk. To run QMT, choose a Windows image instead. PTrade is hosted in the broker's machine room, so this problem does not apply.

09 · Pitfalls

• Download the client only from the broker's channel, and log in with the account the broker issues. A "cracked" build, a service that claims to open the terminal with no threshold, or someone offering to run the strategy for you can end in a banned account, or in the funds and the strategy leaking.
• Paper-trade first. The ThinkTrader community lets you apply for a free 7x24 simulation. For PTrade, ask the broker about a simulated environment. After the backtest passes, run about two weeks of paper trading. The method is paper trading after a backtest.
• A wrong order is your loss. Before you go live, cap the size of one order, the number of orders in a day, and the total position.
• Versions change. xtquant, the built-in Python, and the broker's limits all move. Code that worked can need a change after an update. Watch the ThinkTrader FAQ and the broker's notices.

10 · How to choose

• You write Python and want your own libraries, data, or a model: QMT, with MiniQMT plus xtquant.
• You do not want to mind a computer, and the strategy uses only built-in data and common libraries: PTrade.
• You are still learning and only want a backtest: start with a hosted site such as JoinQuant or RiceQuant. See how to choose a quant trading platform.
• You want the data, the backtest, and the orders on machines you control: an open-source framework such as vn.py. The comparison is how to choose a backtesting framework.

This is not investment advice. Programmatic trading can lose money, and a backtest is not a future return. Features, fees, and opening conditions for QMT and PTrade follow the vendor and the broker.