FinRL®: Financial Reinforcement Learning. 🔥
15.7k
Stars
3.4k
Forks
305
Open issues
30
Contributors
AI Analysis
FinRL is an open-source framework implementing financial reinforcement learning for algorithmic trading research and education. It provides an end-to-end pipeline for training DRL agents on stock market data, benchmarking trading strategies, and prototyping quantitative models. The project targets researchers, students, and developers learning DRL applications in finance—not production traders, for whom the maintainers explicitly direct users to the successor project FinRL-X/FinRL-Trading.
Inferred from signals mentioned in the README (tests, CI, type safety) — not a review of the actual code.
AI's overall editorial judgment — not an average of the bars above, can weigh other factors too.
FinRL: Open-source financial reinforcement learning framework for research and education, now transitioning to FinRL-X
FinRL is the original open-source framework for applying deep reinforcement learning (DRL) to financial trading, covering data ingestion, environment simulation, and agent training in an end-to-end pipeline. It targets researchers, students, and developers exploring algorithmic trading with DRL. With 15,500+ stars, a published arXiv paper, and PyPI downloads tracked publicly, it has genuine academic traction. The project is now explicitly repositioning itself as an educational and benchmarking resource while production focus shifts to its successor, FinRL-X (FinRL-Trading).
Launched in July 2020 by the AI4Finance Foundation, FinRL was among the earliest structured frameworks to apply DRL to quantitative finance. It produced a peer-referenced arXiv paper (2011.09607) and spawned an ecosystem including FinRL-Meta, ElegantRL, and FinGPT.
Growth was initially driven by academic interest in applying reinforcement learning to finance, amplified by the 2020–2021 retail trading boom and ML research community interest. The arXiv paper and Jupyter notebook tutorials lowered the barrier significantly. Growth has plateaued at ~67 stars/week as of mid-2026, consistent with a maturing project in a specialized niche rather than a rapidly expanding one.
PyPI download badge is present and tracked via pepy.tech, indicating measurable install volume, though exact figures are not quoted in the README excerpt. Academic citations via arXiv paper (2011.09607) are documented. Discord community exists. The README explicitly states the framework is for education and research, not production — production use is redirected to FinRL-X. Real-world production trading adoption of the original FinRL library is not directly verified.
Based on the README, FinRL appears to use a three-layer coupled architecture: data layer (14 manually-wired data processors including Yahoo Finance), environment layer (OpenAI Gym-compatible), and agent layer (A2C, DDPG, PPO, SAC, TD3). The README explicitly describes this as a monolithic design, contrasting it with the modular approach of FinRL-X. Jupyter notebooks are the primary delivery mechanism, suggesting a research-oriented rather than production-oriented codebase.
Not documented in README
Last push was 2026-05-26, approximately one month before the evaluation date, indicating active if infrequent maintenance. The README is actively updated to redirect users to FinRL-X. Open and closed issue/PR counts are displayed, suggesting ongoing community engagement. The project appears maintained but in a sustaining mode rather than active feature development.
ADOPT IF: you are a researcher, student, or developer learning DRL applied to finance and want a well-documented, academically grounded starting point with tutorials and benchmarks. AVOID IF: you need production-grade live trading infrastructure, robust risk controls, or a maintained feature roadmap — use FinRL-X instead. MONITOR IF: you want to track the FinRL-X evolution and assess whether the production-oriented successor delivers on its stated modular, deployment-ready architecture.
Independent dimensions
Mainstream potential
3/10
Technical importance
7/10
Adoption evidence
5/10
- Active development focus has shifted to FinRL-X, meaning the original FinRL library is likely in sustaining mode and may fall further behind modern tooling over time.
- Jupyter notebook-centric design limits integration into production Python codebases without significant refactoring.
- Financial RL models trained on historical data are known to overfit and may not generalize to live markets — the framework does not appear to address this risk structurally.
- Dependency on manually-wired data processors (14 as noted in README) may create fragility as data provider APIs change over time.
- The growing dominance of LLM-based trading approaches (e.g., TradingAgents at 88K stars) may reduce academic and practitioner interest in pure DRL frameworks for finance.
FinRL will likely stabilize as a stable educational artifact and benchmark reference, while the AI4Finance Foundation concentrates resources on FinRL-X. Star growth will continue slowly from academic users. Long-term relevance depends on FinRL-X's success.
Newsletter
Get analyses like this every Monday
Free weekly digest of the most interesting open-source discoveries.
Languages
Information
- Website
- https://ai4finance.org
- Language
- Jupyter Notebook
- License
- MIT
- Last updated
- 2mo ago
- Created
- 72mo ago
- Analyzed with
- anthropic/claude-haiku-4-5
Stars over time
Contributors over time
Top 100 contributors only — repos with more will plateau at 100.
Open issues
[Discussion] Connecting factor stock pools with AI / RL trading research
paper_trading/alpaca.py: submitOrder response list never read after thread joins
Add pre-trade market state verification to StockTradingEnv
Feature: Chart pattern similarity as observation/state for RL agents
Top contributors
Recent releases
Similar repos
AI4Finance-Foundation/FinRobot
FinRobot is an open-source AI agent platform designed specifically for...
AI4Finance-Foundation/FinGPT
FinGPT is an open-source large language model framework specifically fine-tuned...
tensortrade-org/tensortrade
TensorTrade is a Python framework for training reinforcement learning agents to...
FareedKhan-dev/all-rl-algorithms
A collection of 18 reinforcement learning algorithms implemented from scratch...
| Repository | Stars | Week Δ | Language | Score | Updated |
|---|---|---|---|---|---|
|
AI4Finance-Foundation/FinRL
this
|
15.7k | +113 | Jupyter Notebook | 7/10 | 2mo ago |
|
AI4Finance-Foundation/FinRobot
|
7.5k | — | Jupyter Notebook | 7/10 | 3d ago |
|
AI4Finance-Foundation/FinGPT
|
20.5k | — | Jupyter Notebook | 8/10 | 1mo ago |
|
tensortrade-org/tensortrade
|
6.4k | — | Python | 7/10 | 5mo ago |
|
FareedKhan-dev/all-rl-algorithms
|
1.8k | — | Jupyter Notebook | 7/10 | 11mo ago |
 RLinf/RLinf
|
4.1k | — | Python | 8/10 | 7h ago |
TradingAgents has 88,546 stars and focuses on LLM-driven trading agents, representing a newer paradigm shift toward language models in finance. It dwarfs FinRL in star count but serves a different technical approach (LLMs vs DRL). Not a direct replacement for DRL-based research.
TensorTrade (6,396 stars) is a closer architectural peer — also a Gym-based trading RL framework in Python. It appears less actively maintained than FinRL and has a smaller ecosystem, but targets a similar developer audience.
FinGPT (20,518 stars) is from the same foundation and focuses on LLMs for finance. More stars than FinRL, reflecting current enthusiasm for LLM-based approaches. Complements rather than competes with FinRL in the AI4Finance ecosystem.
Dopamine (10,875 stars) is a general-purpose RL research framework from Google. Not finance-specific, so it lacks FinRL's domain-adapted environments and financial data pipelines, but may be preferred for users wanting a more rigorous general RL baseline.
rllm (5,646 stars) targets RL with language models, representing another convergence trend. Less mature than FinRL but addresses the emerging LLM+RL intersection, which FinRL-X is also moving toward.
