125B local LLM and Minecraft hit 120 FPS together on an RTX Spark laptop

An enthusiast ran the 125-billion-parameter Qwen 3.8 Flash Next locally on an HP OmniBook Ultra 16 with Nvidia RTX Spark (20-core Arm CPU, 6,144-core Blackwell GPU, 128GB unified memory) while playing Minecraft. A Mindcraft bot controlled a character via in-game chat, and the game held around 120 FPS. The model generated roughly 46 tokens/s at short context and just under 42 at 32,000 tokens.

• Qwen 3.8 Flash Next: 125B parameters, about 6B active per token
• Inference speed: ~46 tokens/s, ~42 tokens/s at 32K context
• Minecraft ran at ~120 FPS alongside the local LLM
• Mindcraft bot on Mineflayer is driven by in-game chat commands