Athene-70B-0725 is a large language model developed by Nexusflow, built on the architecture of Meta's Llama-3-70B-Instruct. It was refined through a specialized post-training pipeline that employs Reinforcement Learning from Human Feedback (RLHF) to optimize performance in conversational tasks, logical reasoning, and instruction following.
Upon its debut in July 2024, the model achieved recognition for its performance on the LMSYS Chatbot Arena, where it demonstrated competitive results against larger proprietary systems. It utilizes the standard Llama 3 configuration, including a context window of 8,192 tokens. The model is part of Nexusflow's Athene series, which focuses on advancing open-weights model capabilities to meet enterprise-grade standards for AI agents and copilot applications.