AReaL

license

Apache-2.0(OSI)

source

public(areal-project/AReaL)

core-gated

ungated(no enterprise or ee path in the repository root and no paid tier described in the README)

Apache-2.0, an OSI licence, with the full source public and the shipped code in this repository. The LICENSE body was read in full: the stock Apache-2.0 text, its core byte-identical to the canonical template, with nothing after END OF TERMS AND CONDITIONS but the standard appendix. No enterprise or ee path in the repository root and no paid tier described in the readme.

• https://raw.githubusercontent.com/areal-project/AReaL/main/LICENSE recorded 2026-08-31

  LICENSE body, read in full: the stock Apache License 2.0 text, carrying no appended term or additional condition.

• https://raw.githubusercontent.com/areal-project/AReaL/main/README.md recorded 2026-08-31

  README, read for the distribution and gating question: it describes installing and running the code from this repository and names no enterprise, paid or licence-gated build.

What changed here is the instrument, not the level. The PyPI package publishes 1.0.4, uploaded 2026-06-10 and the only release the project has ever made, against repository release v2.1.0 of 2026-08-25 - a whole major line, so anyone on the current version reached it another way and a download count off that line measures a superseded release. A usage_volume record claims to be a download count of the product; this one was a download count of a version the project has moved past, which is the wrong claim rather than a wrong number. The record therefore states no count: reported_traction, no numeric reach, and the 45 downloads kept only as corroboration of the trailing channel. The level is deliberately left where the earlier reading put it. Raising it is a separate judgment needing its own standing evidence, and the only other signal available is the star count, which the routing precedence exists to keep out of an adoption band.

• https://pypi.org/pypi/areal/json recorded 2026-08-31

  info.version 1.0.4 with a single entry in `releases`, uploaded 2026-06-10T06:48:31Z. The package has made no other release, so the registry line has never moved past 1.x while the repository is on 2.1.0.

• https://api.github.com/repos/areal-project/AReaL/releases/latest recorded 2026-08-31

  tag_name v2.1.0, published_at 2026-08-25T10:23:59Z, prerelease false - the current stable release line, one major above the sole PyPI release.

• https://pypistats.org/api/packages/areal/recent recorded 2026-08-31

  last_month 45, last_week 9, last_day 3 for the areal package. Recorded as a measurement of the trailing registry line, explicitly NOT as the basis of the band.

• https://api.github.com/repos/areal-project/AReaL recorded 2026-08-31

  Repo metadata for areal-project/AReaL: stargazers_count 5705, archived false, pushed_at 2026-08-31 - the attention figure the note contrasts with the download count.

AReaL's support matrix lists the algorithm, model, training-backend and inference-backend axes that place OpenRLHF at 4: multiple RL algorithms over FSDP/Megatron training with vLLM/SGLang rollout, scaled by TP/PP/EP. It is scored at OpenRLHF rather than above it because the two cover the same ground from different angles - OpenRLHF on Ray-scheduled RLHF at 70B+, AReaL on asynchronous agentic RL - and neither reaches the method catalogue of verl or nemo-rl.

• https://raw.githubusercontent.com/areal-project/AReaL/main/README.md recorded 2026-08-31

  Support Matrix section lists Algorithms, Models, Training Backends and Inference Backends; the README describes a fully asynchronous RL paradigm and carries Math & Reasoning, Agentic RL and Vision-Language Model example sets.