[Roadmap] 2025 Q4 Milestones

[garrett4wade]

AReaL 2025 Q4 Milestone Tracker

Introduction

This document tracks major planned enhancements for AReaL through January 31, 2026. Our development roadmap is organized into two categories to help contributors identify where they can make the most impact:

On-going sections contain features currently under active development by the core AReaL team. These represent our immediate priorities.

Planned but not in progress sections list features with concrete implementation plans that we currently lack bandwidth to pursue. We actively welcome community contributions for these items! If you're interested in contributing to any planned feature, please reach out to discuss implementation details.

---

Backends

On-going

• Single-controller mode [Feature] Add single-controller mode #260
• Detailed profiling for optimal performance across different scales fix: enhance rollout statistics tracking with enqueued state #522 chore: refactor boba GRPO for tracing #527 feat: introduces session-centric tracing APIs #539 etc.
• Low-precision RL training (Megatron FP8)
• Data transfer optimization in single-controller mode
• New PyTorch-native backend: Archon

Planned but not in progress

• Multi-LLM training (different agents with different parameters)
• Auto-scaling inference engines in single-controller mode
• Elastic weight update setup and acceleration
• RL training with cross-node vLLM pipeline/context parallelism

---

Usability

Done

• Add CI pipeline to build Docker images upon release ci: build docker images with GCP #564 ci: automatically tag the dev image upon new releases #574
• Wrap training scripts into trainers
• Refactor FSDP/Megatron engine/controller APIs to finer granularity
• Fully respect allocation mode in trainers/training scripts

On-going

• Flatten the import structure of areal modules

Planned but not in progress

• Support distributed training and debugging in Jupyter notebooks
• Example of using a generative or critic-like reward model
• Support directly constructing inference/training engines without config objects

Canceled

• Rename RemoteSGLang/vLLMEngine as SGLang/vLLMEngine

---

Documentation

Done

• Tutorial on how to write efficient async rollout workflows
• Benchmarking and profiling guide
• Use case guides: offline inference, offline evaluation
• AReaL performance tuning guide
  • Device allocation strategies for training and inference
  • Parallelism strategy configuration for training and inference

Planned but not in progress

• Use case guides: multi-agent training

[aaaandychen]

Hi @garrett4wade , I'm a newcomer and excited to contribute. Could you point me to some good first issues to get started? Any responses will be appreciated！

[garrett4wade]

Hi @aaaandychen , thanks for your contribution in advance! IMO #601 is a good start point. It involves refactoring codes in the TrainEngines. We have a detailed refactoring plan in the issue description. Would you like to give it a try?