A

Areal Boba 2 8B

Developed by inclusionAI
AReaL is an asynchronous reinforcement learning training system developed by Ant Group, designed specifically for large inference models, supporting fast training and cutting-edge performance.
Downloads 1,294
Release Time : 6/3/2025

Model Overview

AReaL is a fully asynchronous reinforcement learning training system designed to help users easily build AI agents, especially good at enhancing the reasoning ability of large language models in mathematics and coding.

Model Features

Asynchronous Reinforcement Learning
Through algorithm-system co-design, support fully asynchronous reinforcement learning to achieve the fastest training speed.
Open and Reproducible
Release all code, datasets, and training recipes to ensure reproducible results.
High Scalability
Adapt to different computing resource settings, and can be seamlessly scaled from a single node to 1K GPUs.
Cutting-edge Performance
Perform excellently in mathematical and coding tasks, and support multi-round agent reinforcement learning.

Model Capabilities

Code Generation
Mathematical Reasoning
Multi-round Dialogue
Reinforcement Learning Training

Use Cases

Programming Assistance
Code Auto-completion
Help developers quickly generate code snippets and improve programming efficiency.
Reach a score of 63.0 on LiveCodeBench v5
Algorithm Competition Problem Solving
Solve programming problems on platforms such as Codeforces.
Reach a score of 1962 (97.5%) on Codeforces
Mathematical Reasoning
Mathematical Problem Solving
Solve complex mathematical problems and proofs.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase