Open Source Python Reinforcement Learning Frameworks
Sort By:
Python Reinforcement Learning Frameworks
View 30 business solutionsBrowse free open source Python Reinforcement Learning Frameworks and projects below. Use the toggles on the left to filter open source Python Reinforcement Learning Frameworks by OS, license, language, programming language, and project status.
-
$300 Free Credits to Build on Google CloudPut your $300 in credit toward real workloads, then keep building with free monthly usage for 20+ products. No commitment and no charge until you upgrade.
-
Ship Agents FasterGemini Enterprise Agent Platform lets you rapidly build, scale, govern and optimize production-ready agents grounded in your organization's data. The platform enables developers to build custom or pre-built agents for virtually any use case. New customers get $300 in free credits.
-
1
DeepSeek R1
Open-source, high-performance AI model with advanced reasoning
DeepSeek-R1 is an open-source large language model developed by DeepSeek, designed to excel in complex reasoning tasks across domains such as mathematics, coding, and language. DeepSeek R1 offers unrestricted access for both commercial and academic use. The model employs a Mixture of Experts (MoE) architecture, comprising 671 billion total parameters with 37 billion active parameters per token, and supports a context length of up to 128,000 tokens. DeepSeek-R1's training regimen uniquely integrates large-scale reinforcement learning (RL) without relying on supervised fine-tuning, enabling the model to develop advanced reasoning capabilities. This approach has resulted in performance comparable to leading models like OpenAI's o1, while maintaining cost-efficiency. To further support the research community, DeepSeek has released distilled versions of the model based on architectures such as LLaMA and Qwen. -
2
DeepSeek-V3
Powerful AI language model (MoE) optimized for efficiency/performance
DeepSeek-V3 is a robust Mixture-of-Experts (MoE) language model developed by DeepSeek, featuring a total of 671 billion parameters, with 37 billion activated per token. It employs Multi-head Latent Attention (MLA) and the DeepSeekMoE architecture to enhance computational efficiency. The model introduces an auxiliary-loss-free load balancing strategy and a multi-token prediction training objective to boost performance. Trained on 14.8 trillion diverse, high-quality tokens, DeepSeek-V3 underwent supervised fine-tuning and reinforcement learning to fully realize its capabilities. Evaluations indicate that it outperforms other open-source models and rivals leading closed-source models, achieving this with a training duration of 55 days on 2,048 Nvidia H800 GPUs, costing approximately $5.58 million. -
3
Brax
Massively parallel rigidbody physics simulation
Brax is a fast and fully differentiable physics engine for large-scale rigid body simulations, built on JAX. It is designed for research in reinforcement learning and robotics, enabling efficient simulations and gradient-based optimization. -
4
TorchRL
A modular, primitive-first, python-first PyTorch library
TorchRL is an open-source Reinforcement Learning (RL) library for PyTorch. TorchRL provides PyTorch and python-first, low and high-level abstractions for RL that are intended to be efficient, modular, documented, and properly tested. The code is aimed at supporting research in RL. Most of it is written in Python in a highly modular way, such that researchers can easily swap components, transform them, or write new ones with little effort. -
Save Up to 91% on Cloud Compute With Spot VMsRun batch jobs at 60-91% off with Spot VMs. Long-running workloads get automatic discounts with sustained use.
-
5
Agent S
Agent S: an open agentic framework that uses computers like a human
Agent S is an open-source agentic framework designed to enable autonomous computer use through an Agent-Computer Interface (ACI). Built to operate graphical user interfaces like a human, it allows AI agents to perceive screens, reason about tasks, and execute actions across macOS, Windows, and Linux systems. The latest version, Agent S3, surpasses human-level performance on the OSWorld benchmark, demonstrating state-of-the-art results in complex multi-step computer tasks. Agent S combines powerful foundation models (such as GPT-5) with grounding models like UI-TARS to translate visual inputs into precise executable actions. It supports flexible deployment via CLI, SDK, or cloud, and integrates with multiple model providers including OpenAI, Anthropic, Gemini, Azure, and Hugging Face endpoints. With optional local code execution, reflection mechanisms, and compositional planning, Agent S provides a scalable and research-driven framework for building advanced computer-use agents. -
6
Machine Learning PyTorch Scikit-Learn
Code Repository for Machine Learning with PyTorch and Scikit-Learn
Initially, this project started as the 4th edition of Python Machine Learning. However, after putting so much passion and hard work into the changes and new topics, we thought it deserved a new title. So, what’s new? There are many contents and additions, including the switch from TensorFlow to PyTorch, new chapters on graph neural networks and transformers, a new section on gradient boosting, and many more that I will detail in a separate blog post. For those who are interested in knowing what this book covers in general, I’d describe it as a comprehensive resource on the fundamental concepts of machine learning and deep learning. The first half of the book introduces readers to machine learning using scikit-learn, the defacto approach for working with tabular datasets. Then, the second half of this book focuses on deep learning, including applications to natural language processing and computer vision. -
7
Gymnasium
An API standard for single-agent reinforcement learning environments
Gymnasium is a fork of OpenAI Gym, maintained by the Farama Foundation, that provides a standardized API for reinforcement learning environments. It improves upon Gym with better support, maintenance, and additional features while maintaining backward compatibility. -
8
ManiSkill
SAPIEN Manipulation Skill Framework
ManiSkill is a benchmark platform for training and evaluating reinforcement learning agents on dexterous manipulation tasks using physics-based simulations. Developed by Hao Su Lab, it focuses on robotic manipulation with diverse, high-quality 3D tasks designed to challenge perception, control, and planning in robotics. ManiSkill provides both low-level control and visual observation spaces for realistic learning scenarios. -
9
Alibi Explain
Algorithms for explaining machine learning models
Alibi is a Python library aimed at machine learning model inspection and interpretation. The focus of the library is to provide high-quality implementations of black-box, white-box, local and global explanation methods for classification and regression models. -
Fully Managed MySQL, PostgreSQL, and SQL ServerCloud SQL handles your database ops end to end, so you can focus on your app.
-
10
Cosmos-RL
Cosmos-RL is a flexible and scalable Reinforcement Learning framework
Cosmos-RL is a scalable reinforcement learning framework designed specifically for physical AI systems such as robotics, autonomous agents, and multimodal models. It provides a distributed training architecture that separates policy learning and environment rollout processes, enabling efficient and asynchronous reinforcement learning at scale. The framework supports multiple parallelism strategies, including tensor, pipeline, and data parallelism, allowing it to leverage large GPU clusters effectively. It is built with compatibility in mind, supporting popular model families such as LLaMA, Qwen, and diffusion-based world models, as well as integration with Hugging Face ecosystems. cosmos-rl also includes support for advanced RL algorithms, low-precision training, and fault-tolerant execution, making it suitable for large-scale production workloads. -
11
DI-engine
OpenDILab Decision AI Engine
DI-engine is a unified reinforcement learning (RL) platform for reproducible and scalable RL research. It offers modular pipelines for various RL algorithms, with an emphasis on production-level training and evaluation. -
12
SLM Lab
Modular Deep Reinforcement Learning framework in PyTorch
SLM Lab is a modular and extensible deep reinforcement learning framework designed for research and practical applications. It provides implementations of various state-of-the-art RL algorithms and emphasizes reproducibility, scalability, and detailed experiment tracking. SLM Lab is structured around a flexible experiment management system, allowing users to define, run, and analyze RL experiments efficiently. -
13
TensorHouse
A collection of reference Jupyter notebooks and demo AI/ML application
TensorHouse is a scalable reinforcement learning (RL) platform that focuses on high-throughput experience generation and distributed training. It is designed to efficiently train agents across multiple environments and compute resources. TensorHouse enables flexible experiment management, making it suitable for large-scale RL experiments in both research and applied settings. -
14
AnyTrading
The most simple, flexible, and comprehensive OpenAI Gym trading
gym-anytrading is an OpenAI Gym-compatible environment designed for developing and testing reinforcement learning algorithms on trading strategies. It simulates trading environments for financial markets, including stocks and forex. -
15
Deep Reinforcement Learning for Keras
Deep Reinforcement Learning for Keras.
keras-rl implements some state-of-the-art deep reinforcement learning algorithms in Python and seamlessly integrates with the deep learning library Keras. Furthermore, keras-rl works with OpenAI Gym out of the box. This means that evaluating and playing around with different algorithms is easy. Of course, you can extend keras-rl according to your own needs. You can use built-in Keras callbacks and metrics or define your own. Even more so, it is easy to implement your own environments and even algorithms by simply extending some simple abstract classes. Documentation is available online. -
16
H2O LLM Studio
Framework and no-code GUI for fine-tuning LLMs
Welcome to H2O LLM Studio, a framework and no-code GUI designed for fine-tuning state-of-the-art large language models (LLMs). You can also use H2O LLM Studio with the command line interface (CLI) and specify the configuration file that contains all the experiment parameters. To finetune using H2O LLM Studio with CLI, activate the pipenv environment by running make shell. With H2O LLM Studio, training your large language model is easy and intuitive. First, upload your dataset and then start training your model. Start by creating an experiment. You can then monitor and manage your experiment, compare experiments, or push the model to Hugging Face to share it with the community. -
17
RL Baselines3 Zoo
Training framework for Stable Baselines3 reinforcement learning agents
rl-baselines3-zoo is a collection of pre-trained models, benchmarks, and hyperparameter tuning tools built on top of Stable Baselines3, a reinforcement learning library. It provides an easy way to test, evaluate, and train RL agents across a wide variety of environments. -
18
verl
Volcano Engine Reinforcement Learning for LLMs
VERL is a reinforcement-learning–oriented toolkit designed to train and align modern AI systems, from language models to decision-making agents. It brings together supervised fine-tuning, preference modeling, and online RL into one coherent training stack so teams can move from raw data to aligned policies with minimal glue code. The library focuses on scalability and efficiency, offering distributed training loops, mixed precision, and replay/buffering utilities that keep accelerators busy. It ships with reference implementations of popular alignment algorithms and clear examples that make it straightforward to reproduce baselines before customizing. Data pipelines treat human feedback, simulated environments, and synthetic preferences as interchangeable sources, which helps with rapid experimentation. VERL is meant for both research and production hardening: logging, checkpointing, and evaluation suites are built in so you can track learning dynamics and regressions over time. -
19
CleanRL
High-quality single file implementation of Deep Reinforcement Learning
CleanRL is a Deep Reinforcement Learning library that provides high-quality single-file implementation with research-friendly features. The implementation is clean and simple, yet we can scale it to run thousands of experiments using AWS Batch. CleanRL is not a modular library and therefore it is not meant to be imported. At the cost of duplicate code, we make all implementation details of a DRL algorithm variant easy to understand, so CleanRL comes with its own pros and cons. You should consider using CleanRL if you want to 1) understand all implementation details of an algorithm's variant or 2) prototype advanced features that other modular DRL libraries do not support (CleanRL has minimal lines of code so it gives you great debugging experience and you don't have to do a lot of subclassing like sometimes in modular DRL libraries). -
20
Google Research Football
Check out the new game server
Google Research Football is a reinforcement learning environment simulating soccer matches. It focuses on learning complex behaviors such as team collaboration and strategy formation in competitive settings. -
21
Habitat-Lab
A modular high-level library to train embodied AI agents
Habitat-Lab is a modular high-level library for end-to-end development in embodied AI. It is designed to train agents to perform a wide variety of embodied AI tasks in indoor environments, as well as develop agents that can interact with humans in performing these tasks. Allowing users to train agents in a wide variety of single and multi-agent tasks (e.g. navigation, rearrangement, instruction following, question answering, human following), as well as define novel tasks. Configuring and instantiating a diverse set of embodied agents, including commercial robots and humanoids, specifying their sensors and capabilities. Providing algorithms for single and multi-agent training (via imitation or reinforcement learning, or no learning at all as in SensePlanAct pipelines), as well as tools to benchmark their performance on the defined tasks using standard metrics. -
22
TextWorld
TextWorld is a sandbox learning environment for the training
TextWorld is a learning environment designed to train reinforcement learning agents to play text-based games, where actions and observations are entirely in natural language. Developed by Microsoft Research, TextWorld focuses on language understanding, planning, and interaction in complex, narrative-driven environments. It generates games procedurally, enabling scalable testing of agents’ natural language processing and decision-making abilities. -
23
VectorizedMultiAgentSimulator (VMAS)
VMAS is a vectorized differentiable simulator
VectorizedMultiAgentSimulator is a high-performance, vectorized simulator for multi-agent systems, focusing on large-scale agent interactions in shared environments. It is designed for research in multi-agent reinforcement learning, robotics, and autonomous systems where thousands of agents need to be simulated efficiently. -
24
AgentUniverse
agentUniverse is a LLM multi-agent framework
AgentUniverse is a multi-agent AI framework that enables coordination between multiple intelligent agents for complex task execution and automation. -
25
Atropos
Language Model Reinforcement Learning Environments frameworks
Atropos is a comprehensive open-source framework for reinforcement learning (RL) environments tailored specifically to work with large language models (LLMs). Designed as a scalable ecosystem of environment microservices, Atropos allows researchers and developers to collect, evaluate, and manage trajectories (sequences of actions and outcomes) generated by LLMs across a variety of tasks—from static dataset benchmarks to dynamic interactive games and real-world scenario environments. It provides foundational tooling for asynchronous RL loops where environment services communicate with trainers and inference engines, enabling complex workflow orchestration in distributed and parallel setups. This framework facilitates experimentation with RLHF (Reinforcement Learning from Human Feedback), RLAIF, or multi-turn training approaches by abstracting environment logic, scoring, and logging into reusable components.
