Freqtrade review (2026)

Freqtrade is the default answer to "how do I backtest and run a crypto trading bot for free in 2026," and the project's momentum backs that up: 52.9k GitHub stars, calendar-versioned releases shipping roughly monthly, and an 11k-member Discord. Unlike Backtrader or Zipline Reloaded, this is not a maintenance-mode library; it is one of the most actively developed open-source trading projects in existence, and every feature, including the FreqAI machine-learning module, ships free under GPL-3.0 with no paid tier.

The scope is narrower than a general-purpose backtester, though. Freqtrade only trades crypto, bundles no data of its own (you download candles from whichever exchange you configure), has no native walk-forward module, and its biggest headline feature, Hyperopt, is also its biggest risk: the documentation itself warns that over-precise parameter searches produce overfitted results, and the GitHub issue tracker has years of reports of Hyperopt and backtest numbers disagreeing on the same inputs.

Our verdict: for crypto-only strategy development with a real path from research to live automation, Freqtrade is the strongest free option we cover, ahead of general Python libraries like Backtrader (6.8/10) on maintenance and multi-pair portfolio support. We score it 7.0/10. If you need equities, forex or futures, or a formal walk-forward workflow, look at NautilusTrader (7.9) or QuantConnect (8.3) instead.

Freqtrade is a free, open-source Python framework for backtesting, optimizing and running automated cryptocurrency trading strategies, first released in 2017 and still under active development. Unlike libraries such as Backtrader or Backtesting.py, which are pure research tools, Freqtrade is built as a complete bot: the same strategy class you backtest can run in dry-run (simulated live) mode or execute real orders, controlled and monitored through a Telegram bot, a REST API, or the FreqUI web dashboard.

Strategies are Python classes that compute indicators over pandas DataFrames and emit entry/exit signals, similar in spirit to Backtesting.py or vectorized Pine Script logic. Around that core sit the pieces that make it a full product: a pair-whitelist system for testing and trading multiple markets from one config, a data downloader built on the CCXT library, Hyperopt for parameter search, and FreqAI for layering machine-learning models on top of a strategy's feature set.

The project counts 52.9k GitHub stars, over 32,500 commits, and 11k forks as of mid-2026, with releases on a calendar-based versioning scheme (2026.4 through 2026.7 shipped April through July 2026 alone). That release cadence, sustained since 2017, is a sharp contrast to the frozen or slow-moving state of several older Python backtesting libraries.

Freqtrade's backtester replays historical OHLCV candles sequentially rather than vectorizing the whole series at once, evaluating entry and exit conditions candle by candle in a way that mirrors how the live bot actually trades. By default it assumes trades enter at the next candle's open and fills orders at the requested price with no slippage, as long as that price sits within the candle's high/low range; ROI-based exits reference the candle's high, and stoploss exits trigger at the exact stop price. The documentation is upfront that this loses information about what happens inside a candle, and offers --timeframe-detail to re-simulate fills using a lower timeframe (e.g., checking 1-minute data inside 1-hour candles) for more realistic sequencing. That flag is opt-in, so a default backtest without it should be read as a rough pass, not a final answer.

Multi-pair handling is a genuine strength versus most Python backtesting libraries: a single backtest run manages a configurable pair whitelist under a shared max_open_trades cap and supports dynamic stake sizing with profit compounding, which is closer to real portfolio behavior than the single-instrument-per-run model of Backtrader or Backtesting.py. Position reversals are supported for futures pairs when an opposing signal fires on the same candle.

The weak point is Hyperopt, the built-in Bayesian optimizer (currently Optuna's NSGA-III sampler under the hood). It is powerful and genuinely useful for narrowing a search space, but Freqtrade's own docs warn that decimal precision beyond 0.001 'will usually result in overfitted results,' and the GitHub issue tracker has a long, recurring thread of reports where Hyperopt's reported performance does not reproduce when the same parameters are run back through the standard backtest command. Treat every Hyperopt result as a hypothesis to re-validate, not a finished strategy, exactly as the maintainers themselves advise.

Freqtrade ships with zero bundled data. The built-in freqtrade download-data command pulls historical candles directly from whichever exchange you configure, via the CCXT library, which in principle reaches 100+ exchanges. The project officially tests and documents quirks for a smaller, curated list: Binance, Kraken, OKX, Bybit, Gate.io, Bitget, HTX, Bitmart, Bitvavo, Kucoin, and the DEX-based Hyperliquid, each with exchange-specific notes (Kraken's 720-candle historical limit, Kucoin and OKX requiring an API passphrase, Bitmart needing a verification tier, and so on).

History depth and resolution are entirely dictated by what the exchange's public API exposes; there is no independent, survivorship-bias-corrected dataset behind it the way QuantConnect or TradeStation bundle for equities and futures. Resolution tops out at 1-minute candles; there is no tick or order-book data. And critically, this is a crypto-only tool: no stocks, ETFs, forex or traditional futures, in exchange for treating spot and (on supported exchanges) perpetual futures as first-class citizens with proper funding-rate and liquidation awareness in live trading.

For a crypto-focused quant, the download command's tight CCXT integration and exchange-specific handling of quirks (rate limits, historical limits, futures margin modes) is meaningfully more turnkey than Backtrader's bitrotted Yahoo loader or Backtesting.py's total silence on data sourcing. For anyone trading other asset classes, this section is a hard disqualifier.

There is no pricing page because there is no product to buy. Freqtrade is licensed GPL-3.0 and distributed via GitHub and PyPI at no cost, and we verified in August 2026 that the FreqAI documentation states outright it is 'and always will be, a not-for-profit, open source project' with no crypto token and no paid signal service. The project sustains itself through volunteer maintenance and individual GitHub Sponsors pages for core contributors, not through a commercial tier of the software itself.

The FAQ contains a pointed warning worth repeating here: because the project is popular and free, scam sites periodically advertise a 'Freqtrade token' or an official crypto offering. The maintainers state clearly that no such token exists and that any site claiming one is fraudulent. That is a real, documented risk in the ecosystem around the tool, not a knock on the tool itself.

Your actual costs are indirect: exchange trading fees on live orders, API rate limits that can slow large data downloads, and, if you want the bot running continuously rather than on your own machine, a small VPS (commonly a few dollars a month). None of that is charged by the Freqtrade project.

This is the section where Freqtrade separates itself from most of the free Python backtesting field. Where Backtrader has had no release since April 2023 and Zipline Reloaded moves slowly, Freqtrade ships on a calendar-based release train, with four releases (2026.4 through 2026.7) landing between April and July 2026 alone, each with real changelog entries (the July release, for example, improved backtesting performance for --export signals and added Arrow-based filtering for feather/parquet OHLCV data). The GitHub repository shows over 32,500 commits and an active issue tracker where maintainers respond to bug reports, including the Hyperopt-versus-backtest discrepancies discussed above.

Community size backs this up: 52.9k GitHub stars, 11k forks, and a Discord server with roughly 11,000 members where users trade strategy ideas, debug configurations, and discuss FreqAI feature engineering. A companion freqtrade-strategies repository and a wider ecosystem of community strategy repos on GitHub give newcomers real starting points, in contrast to needing to build every strategy from a blank file.

The honest caveat is that Freqtrade's active development does not immunize it against the core risk of retail algo trading: strategies that look good in Hyperopt or backtest can still fail live, and the volume of GitHub issues about hyperopt/backtest mismatches suggests this trips up a meaningful share of users, not just careless ones.

Freqtrade fits three groups well. First, Python-comfortable crypto traders who want to backtest and automate strategies without paying for a platform: the combination of free multi-pair portfolio backtesting, Hyperopt, and direct live deployment to 10+ exchanges is hard to match at any price. Second, people specifically interested in applying machine learning to trading, since FreqAI is a mature, actively maintained ML layer that would otherwise require significant custom engineering on top of a generic backtester. Third, anyone who values an actively maintained dependency: the release cadence and community size are real risk-reduction versus adopting a frozen library.

It is a poor fit for traders who need equities, forex, or traditional futures (crypto only, full stop), anyone who wants formal walk-forward analysis without building it themselves, and beginners who are not ready to read Python source code when something breaks, since the project's own support culture leans on self-sufficiency. If your universe is crypto and you can write Python, Freqtrade is one of the strongest free options available in 2026. If it is not, this tool has nothing to offer you.