量子位官网T2·78 pts
- Gamer is trained from the 8B-parameter Qwen3-VL-8B and maps screen pixels plus task instructions directly to keyboard and mouse actions without game APIs or internal state
- It ran autonomously for over 95% of a roughly 50-hour Star Rail playthrough, handling boss fights, environmental mechanisms and NPC-following navigation
- Its key contribution is unifying cross-game real-time control with long-horizon memory and planning in one VLA framework that tracks completed and upcoming steps