[Roadmap] 2026 Q1 Milestones

[garrett4wade]

AReaL 2026 Q1 Milestone Tracker

Introduction

This document tracks major planned enhancements for AReaL through April 30, 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

• ZBPP & ZBPP-V support for the Archon backend feat(archon): add InterleavedZeroBubble (ZB1P) pipeline schedule #936 feat(archon): add ZBVZeroBubble pipeline schedule support #916
• FP8 training for Archon
• Online RL training with the proxy server feat(proxy): add proxy gateway and online RL training mode #947

Planned but not in progress

• Support for agentic training with large VLM MoE models (Archon backend)
• Omini model RL support with FSDP/Archon backend [Feature] Support Audio model in the future #879
• Decoupling agent service with the inference service
• LoRA support for the Archon backend
• Colocation mode with awex as the weight sync engine
• 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

On-going

• Flatten the import structure of areal modules

Planned but not in progress

• Publishing pypi packages
• 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
• Add router in rollout controller for simpler proxy server usage
• Integrating aenvironment for environment handling

---

Documentation

On-going

N/A

Planned but not in progress

• Use case guides: multi-agent training
• Guide for online proxy mode training

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity within the last 14 days.

If this issue is still relevant, please add a comment to keep it open. Otherwise, it will be automatically closed in 16 days.

Thank you for your contribution!

[astrajwt]

Hi! I’m new here and very interested in contributing to AReaL.

Are there any beginner-friendly tasks I could help with, such as documentation improvements or small bug fixes?

I noticed in the roadmap that multimodal support (especially audio/VLM models) is planned. I have previous experience adapting vision and audio models for training/inference pipelines. I’m curious whether the community has already started working on this area, and if there’s any way I could contribute.

Thanks a lot!

[rchardx]

Hi! I'm new here and very interested in contributing to AReaL.

Are there any beginner-friendly tasks I could help with, such as documentation improvements or small bug fixes?

I noticed in the roadmap that multimodal support (especially audio/VLM models) is planned. I have previous experience adapting vision and audio models for training/inference pipelines. I'm curious whether the community has already started working on this area, and if there's any way I could contribute.

Thanks a lot!

Hi @astrajwt, welcome to AReaL! Glad to have you here 🎉

Multimodal / VLM — we could really use your help here!

We do have early VLM support landed — Qwen2.5-VL and Qwen3-VL under the FSDP engine, with existing examples in examples/vlm/ (Geometry3K, CLEVR Count 70K) and a VisionRLVRWorkflow in areal/workflow/vision_rlvr.py.

That said, the current VL implementation has significant room for optimization, and one high-priority task right now is verifying and fixing the VL workflow end-to-end, starting from the FSDP engine. The current implementation needs thorough validation — correctness of vision-language data flow through training, proper image processing, weight update handling for multimodal models, etc. A great reference point is the NPU VLM examples under examples/vlm_npu/, which have a working pipeline that can serve as a blueprint for what the GPU/FSDP path should look like.

Regarding audio — we do plan to introduce audio modality support in the future. Once the VL workflow is solid, extending the multimodal pipeline to audio models would be a natural next step. Your experience in this area would be very valuable when we get there.

Given your background with vision and audio model pipelines, this would be a highly impactful contribution. If you're interested, feel free to start by running the existing VLM examples, cross-referencing with the NPU examples, and opening a draft PR with your findings. We're happy to provide guidance along the way.

Looking forward to your contributions! 🙌

[astrajwt]

Gotcha — happy to take this on! 💪

[Zijun9]

Hi! I’m new to AReaL and very interested in contributing.

I noticed in the roadmap that there are several planned features such as Agent / Multi-LLM training. I would be happy to explore contributing in that direction, or help with other related areas if there are tasks that need support.

Please feel free to let me know if there are any features or issues where I could help. I’d be glad to get involved!

Thanks!

[rchardx]

Hi! I'm new to AReaL and very interested in contributing.

I noticed in the roadmap that there are several planned features such as Agent / Multi-LLM training. I would be happy to explore contributing in that direction, or help with other related areas if there are tasks that need support.

Please feel free to let me know if there are any features or issues where I could help. I'd be glad to get involved!

Thanks!

Hi @Zijun9, welcome to AReaL! Great to have you here! 🎉

We're always excited to have new contributors. Feel free to explore any area that interests you — whether it's the Agent/Multi-LLM training features you mentioned, or any other part of the roadmap under "Planned but not in progress". All contributions are welcome!

If you encounter any issues, rough edges, or have ideas for improvement while trying out AReaL, feel free to open an issue or submit a PR directly.

We're happy to provide guidance along the way. Looking forward to your contributions! 🙌