OSRS Turned Into an RL Benchmark Environment
jsuarez · x · 2026-07-16
The author shares their work porting OSRS into a high-performance C RL environment to serve as a research benchmark for future algorithmic updates. They demonstrated a small model defeating Sol Heredit with an 87% win rate after training for just 15 minutes on a single 4090 GPU without scaffolding.
The post also clarifies:
- This is strictly for research purposes, not for creating bots for the live game.
- The task is easier than the Inferno boss, aligning with the community sentiment that the real Inferno is harder than the Colosseum.
- The observation vector contains about 3000 floats, primarily derived from items and NPCs.
- A flat MLP performed poorly initially. Switching to an entity-shared MLP with max pooling per group instantly achieved SOTA, consistently outperforming the baseline during hyperparameter sweeps.
More from Research
- New paper defines self-state attacks, showing OS defenses leave four agent-memory cases indistinguishable — Justgototheeffinmoon · 2026-07-22
- Krea 2 users recommend a two-pass Clownshark sampler setup for sharper image details — listopalafoto · 2026-07-22
- Animation shows how an MLP’s first-layer weights change while learning MNIST — CatAstro_Piyush · 2026-07-22
- Project APE finds verifier reliability drops when papers contain multiple errors — soumitrashukla9 · 2026-07-22
- Project APE says verifier costs fell about 90x in a year as Chinese open models lead — soumitrashukla9 · 2026-07-22
- OpenAI-linked paper says capability RL can make models more reward-seeking — MariusHobbhahn · 2026-07-22