PufferLib 5.0 RL Results: 19.5k NetHack Score with Random Classes, Multitask Drone Policies
j_foerst · x · 2026-09-14
jsuarez published a rundown of the five coolest RL results in PufferLib 5.0, with @rockt highlighting a 19.5k score in the NetHack Learning Environment using random classes, regularly reaching The Castle.
Highlights shared in the thread:
- Multitask drone policies: a single policy that can both race and arrange into several different formations.
- Robocode champions via self-play: the 2010 tank-programming game and its monster bots were ported to a high-performance C PufferEnv; PufferLib 5.0 can crush any bot it trains against.
The full list spans five results, showcasing PufferLib 5.0's RL training capability in high skill-ceiling environments.
Related event: PufferLib 5.0 Sets New RL Records on Craftax and NetHack(4 posts)→
More from Research
- Spatial Hub aggregates ~5000 public COLMAP-compatible spatial datasets for world models — janusch_patas · 2026-10-31
- ARC Prize 2026 closes in 3 days as Yi-Chia Chen tops ARC-AGI-3 at 75.56% — GregKamradt · 2026-10-31
- NVIDIA Nemotron 3.5 scales LLM self-verification, +9-10 points on agentic benchmarks — sarahcat21 · 2026-10-31
- Unofficial OpenTargets MCP Server Brings Gene-Drug-Disease Data to LLMs — modelcontextprotocol · 2026-10-31
- Researchers find 5 personality structures in LLM weight space, Big Five-like — luke_drago_ · 2026-10-31
- NVS benchmarks are misaligned: proper pose alignment puts 3D Gaussians back at SOTA — kwangmoo_yi · 2026-10-31