Pi Agent Framework and Qwen 3.8 Flash Next Autonomously Play Baldur's Gate 2
A developer has demonstrated an AI agent setup using the Pi agent framework and the Qwen 3.8 Flash Next language model autonomously playing the classic role-playing game Baldur's Gate 2 on a live stream. The creator also announced plans to open-source the custom integration code powering the stream in the future. Navigating complex classic role-playing games tests an LLM's capacity for long-horizon planning, context retention, and dynamic action selection beyond standard static benchmarks. This project demonstrates the practical potential of pairing lightweight, efficient open-weights models with agent toolkits to interact with unscripted game environments. The setup utilizes Qwen 3.8 Flash Next—a Mixture-of-Experts (MoE) architecture designed for efficient inference—alongside the Pi agent framework to manage state loops and tool execution. While the live playthrough was demonstrated on YouTube, technical implementation details and code repositories remain unreleased for now.
## BACKGROUND
Baldur's Gate 2 is a complex classic computer role-playing game featuring non-linear narrative choices, intricate inventory management, and tactical combat, making it a challenging environment for autonomous AI agents. AI agent frameworks like Pi agent provide the core scaffolding—handling message routing, memory management, and tool calls—necessary for an LLM to take actions in external applications. Qwen 3.8 Flash Next is an open-weights Mixture-of-Experts language model created by Alibaba's Qwen team, designed to balance high performance with lower computational cost.