Freqtrade vs vectorbt vs Backtrader for crypto backtests
Freqtrade, vectorbt, Backtrader and QuantConnect LEAN compared from their own docs: crypto data, live trading, licences, limits, and a look-ahead trap.
Pick Freqtrade if you want a ready-made crypto bot that downloads exchange candles and backtests them. Pick vectorbt if you want to test many ideas fast in Python. Backtrader is a Python framework with documented live trading through Interactive Brokers and Oanda. QuantConnect LEAN fits people who want one engine for many asset types, with crypto data from several exchanges.
This comparison uses each project's own documentation. It also covers a risk no tool removes for you: a timing bug can make a losing test look like a winner.
What is each tool, in plain words?
A backtest runs a trading rule on old prices to see what it would have done. An open-source tool has public code that you can run yourself. A licence is the legal terms for using it.
| Tool | What its docs call it | Licence | Price |
|---|---|---|---|
| Freqtrade | A free open-source Python crypto trading bot with backtesting, plotting, a dry-run mode, strategy optimization and a Telegram and web interface | GPL-3.0 | Free; no paid plans on the pages read |
| vectorbt | A Python package for analysis and backtesting on pandas and NumPy, sped up with Numba and optional Rust code | Apache 2.0 with Commons Clause | Free open-source package; VectorBT PRO starts at $25 a month, $240 for 12 months or $500 once |
| Backtrader | A feature-rich Python framework for backtesting and trading | GPL-3.0 | Free; no paid plans on the pages read |
| QuantConnect LEAN | An open-source, event-driven trading engine for research, backtesting, optimization and live trading, in Python and C# | Apache 2.0 | Engine can run on-premises; the hosted Free Plan has no dollar price, and paid plans are shown without prices |
The Commons Clause in vectorbt's licence means selling products or services mainly based on it is not allowed without an exception. Both Freqtrade and vectorbt carry notes that their software is for educational use.
What do they say about crypto data and live trading?
| Tool | Crypto data, as documented | Live trading, as documented |
|---|---|---|
| Freqtrade | Downloads candles per exchange and timeframe through CCXT. Binance and Hyperliquid are listed. Futures mode also pulls mark and funding-rate candles | Runs as a trading bot, with a dry-run mode |
| vectorbt | Examples use Yahoo Finance. PRO lists data connectors, with no exchanges named on the page read | Not documented in the pages read |
| Backtrader | Feeds for CSV, Pandas and Yahoo. A community recipe called bt-ccxt-store is listed | Interactive Brokers, Visual Chart and Oanda |
| QuantConnect LEAN | Crypto datasets for Binance, Binance US, Bitfinex, Bybit and Coinbase, from ticks to daily bars | Yes, but the Free Plan has 0 live trading nodes |
CCXT is a free library that connects to many exchanges with one interface.
On speed, vectorbt is the one that says so: it packs many strategies, assets and periods into arrays to test them together. It also says large jobs are limited by memory. We did not time any of these tools.
On research checks, the free vectorbt edition lists walk-forward optimization, which tests a rule on later dates than it was tuned on. VectorBT PRO also lists an MCP server for coding agents. Freqtrade's backtest report shows trades, fees-included profit, drawdown, profit factor and Sharpe.
No official page we read measures learning curve, so we do not rank it. Try each tutorial with a small test first.
What limits do their own docs state?
Fairness means listing the limits too.
- Freqtrade. Its docs say backtesting cannot see price moves inside a candle. It assumes entries at the open, no slippage, and a stop-loss checked before the profit target. Data only goes back as far as the exchange supplies, and for Hyperliquid that is about 5000 candles. The default download is the last 30 days of 1-minute and 5-minute candles unless you ask for more. Binance open interest history is limited to roughly 30 days.
- vectorbt. The free edition lacks features reserved for PRO, including parallelization, extra data integrations, portfolio optimization, pattern recognition and limit orders.
- Backtrader. The official site shows its community forum as disabled and points to Stack Overflow. We did not find a maintenance statement on the pages read. Native crypto exchange feeds are not documented.
- QuantConnect LEAN. The Free Plan offers unlimited backtesting but has 1 backtest node, 1 research node and no paper or live trading. Start dates of the crypto history were not stated on the page read.
The words "not documented" mean we did not find a statement. They do not prove that a feature is missing.
Where does Comparand fit?
Comparand is not a backtesting engine. It does not run your rule or place orders. It gives you price data and past look-alike charts to use in whatever tool you pick.
- Free bars. venue_bars returns open, high, low, close and volume for completed bars, from 1-minute to 1-day, on dozens of exchanges. It is free on every key.
- Whole-file catalog. hl_bulk_files lists whole-day files from the Hyperliquid archive (1-minute bars, 5-minute bars, labels, cascade follow-up data, plus funding per coin) with sizes and checksums. That archive covers Hyperliquid only.
- Past look-alikes. shape_analogs finds past charts that look like the current one, for the same coin on the same exchange, and shows what followed each. None of the reviewed official pages for these four tools documents this kind of search.
Backtrader's docs list CSV and Pandas feeds, and vectorbt works on pandas tables, so bars you fetch can be loaded into either once saved in that form.
What lesson applies to every backtester?
Any tool runs the rule you give it. If the rule reads information too early, the result is fake.
We tested one example on Binance BTCUSDT hourly prices, from September 2019 to September 2026: 61,901 hours, with a 0.05% fee per trade side. The honest rule used the last finished hour to pick a position for the next hour. It averaged a loss of 0.06% per hour, with a Sharpe score of −8.00. Sharpe compares the average result with how much results swing.
The broken rule used the direction of the hour it was about to trade. It averaged a gain of 0.33% per hour, with a Sharpe of 59.41. A rule that enters at a candle's open can only use candles that finished before it. We did not test whether each tool above warns about this bug.
Three checks work in any tool. First, write down when each input was first known and compare that with when the trade starts. Second, delete all prices after one decision time and rerun that decision; it should not change. Third, make the rule use data one bar older and see whether the result collapses. Read the full example in look-ahead bias in a backtest, and compare other approaches in ways to backtest a crypto chart pattern.
Fees matter in every tool too. Check which fee is set in yours; trading fees in a backtest shows how much they can change a result.
Which tool fits which person?
- You want a crypto bot with dry-run testing: Freqtrade is built for that, and pulls exchange candles itself.
- You want to compare many ideas quickly in Python: vectorbt describes exactly that. Check the PRO-only list first.
- You want a Python framework with documented broker links: Backtrader names Interactive Brokers and Oanda. Check how crypto data would get in.
- You want one engine for stocks, forex, options and crypto: LEAN lists all of them. Check the plan limits.
This is a guide to fit, not a ranking, and not advice to trade.
How we compared them
We read each project's own site, README and documentation, and used only what those pages state. We did not run or time any of the four tools. Prices are what the pages showed, and can change. The look-ahead numbers come from our own hourly Bitcoin test, which charged 0.05% per trade side and left out funding payments, spread and slippage.
Try this in Comparand
Ask your connected AI agent: "Get completed hourly bars for Binance BTCUSDT between [start] and [end] UTC. Tell me if any hours are missing."
- venue_bars returns open, high, low, close and volume for completed bars only. It is a free data source for any backtest. Your agent checks gaps and saves a file you can load.
Then ask: "List the Hyperliquid archive files for [a month] and show their sizes."
- hl_bulk_files is a free catalog of whole-day files, 100 days per page. It lists files; it does not download them.
- To download, hl_bulk_link gives signed links that last 24 hours and allow 5 downloads each. Plans include 40, 200 or 800 GB of links a month (Starter, Pro, Desk). Pay-as-you-go pays $0.15 per GB.
See the docs to connect, and try 3 coins free to see past look-alike charts. Those are past cases, not a forecast.
The short version
- Freqtrade fits crypto bot builders, vectorbt fits fast Python research, Backtrader is a Python framework with documented broker links, and LEAN covers many asset types.
- Each project's own docs state limits: Freqtrade's candle depth, vectorbt's PRO-only features, Backtrader's undocumented crypto feeds and LEAN's plan nodes.
- None of the official pages reviewed for these four tools documents a historical chart look-alike search.
- A timing bug turned a Bitcoin rule that lost 0.06% per hour into one that seemed to gain 0.33% per hour.
- Use free bars from venue_bars, and check timing in any tool before you trust a result.
Questions people ask
Is Freqtrade or vectorbt better for crypto backtests?
It depends on the job. Freqtrade is a Python crypto trading bot that downloads exchange candles, backtests and has a dry-run mode. vectorbt is a Python package for testing many ideas at once on pandas and NumPy data. Pick by the job, not by a ranking.
Are Freqtrade, vectorbt and Backtrader free?
Freqtrade and Backtrader are open source under GPL-3.0. The open-source vectorbt package is free under Apache 2.0 with Commons Clause. Its paid VectorBT PRO page shows plans starting at $25 per month, $240 for 12 months, or $500 once.
Does Backtrader support crypto exchanges?
Its documentation names Interactive Brokers, Visual Chart and Oanda for live trading, and no crypto exchanges in the README. The docs list a community recipe called bt-ccxt-store, with no details. Native crypto feeds are not documented.
Do these tools protect against look-ahead bias?
We did not test that. The bug is in how a rule reads its data. In our Binance Bitcoin test, one timing bug turned a rule that lost 0.06% per hour after fees into one that seemed to gain 0.33% per hour.