Reinforcement learning frameworks & training environments: top picks for 2026

Open-source desktop app and web UI for running, fine-tuning and exporting LLMs and diffusion models locally, plus a code-based training library.

Unified framework for efficiently fine-tuning 100+ LLMs and VLMs with a zero-code CLI and Web UI.

AI-oriented quantitative investment platform from Microsoft covering data, model training, backtesting and order execution for quant research.

Ray is a unified framework for scaling Python and AI applications from a laptop to a cluster, with a distributed runtime and AI libraries.

verl is an open-source reinforcement learning training library for post-training large language models, built on the HybridFlow paper.

Unity ML-Agents Toolkit lets Unity games and simulations serve as environments for training AI agents with reinforcement and imitation learning.

A Hugging Face library for post-training transformer language models with SFT, GRPO, DPO, KTO and reward modeling.

A lightweight agentic RL framework from Microsoft for training existing AI agents with reinforcement learning while they keep their real tools and harnesses.

An open-source Python framework for applying deep reinforcement learning to automated stock trading and portfolio tasks.

A fine-tuning, training, inference and deployment framework for 600+ large language models and 400+ multimodal models, covering SFT, DPO and GRPO.

A general-purpose physics engine for fast, accurate simulation of articulated bodies in contact, widely used in robotics and machine learning research.

PyTorch implementations of reinforcement learning algorithms, designed as a reliable, consistent baseline for research and applications.

A standard Python API for single-agent reinforcement learning environments, plus a set of reference environments, forked from OpenAI's Gym.

A free, open-source framework for fine-tuning and post-training large language models, configured through YAML files.

A from-scratch PyTorch walkthrough and codebase for training a transformer LLM, covering pretraining, SFT, reward modeling, DPO/PPO/GRPO and generation.

Python RL framework that trains multi-step LLM agents on their own experience using GRPO, with a client-server design that plugs into existing apps.

A Ray- and vLLM-based framework for reinforcement learning from human feedback and agentic RL training of large language and vision-language models.

Open-source Rust software that runs a tiny biped duck robot, covering motion control, gamepad, updates and camera streaming.

A framework from the Terminal-Bench team for evaluating and optimizing AI agents and language models in container environments.

An open-source framework for training LLM agents with reinforcement learning, supporting any agent harness, sandbox, and training backend.

A large-scale asynchronous reinforcement learning system for training LLM reasoning and agentic models, with RL training for black-box agents via a base_url swap.

A bilingual collection of 100+ original diagrams explaining LLM, RL, RLHF, DPO, GRPO and related large-model algorithms.

A JAX-based open-source LLM library and reference implementation for pre-training and post-training large models on Google Cloud TPUs and GPUs.

A short introductory book on reinforcement learning, from basics to applied algorithms, with PyTorch implementations in the repo.

An open framework for building embodied agents that use LLM planners, VLA manipulation models and memory to run robot tasks in simulators and on real robots.

A research framework that uses world-model predictions and interaction feedback to drive recursive self-improvement across code repair, GUI tasks and lab planning.

An open-source reinforcement learning agent for competitive Guandan, a four-player partnership card game, with a playable web table.

ASPIRE is a framework where coding agents iteratively refine reusable robot skills for simulated and real-robot manipulation tasks instead of training model weights.

A Go-based Fight the Landlord card game with online multiplayer, reconnection, smart bots, and a fully random deal with no card-control algorithms.

A Python simulator for UAV swarm communication networks, covering routing, MAC, physical layer, mobility and energy models.

Turns any Agent Skill into RL-ready terminal tasks in the Harbor format, using Codex agents inside Docker containers.