PatronusAI open-sources GlobeBench: 1,632 RL-environment tasks, only GPT-6 Astra tops 30%
cephaloform · x · 2026-11-07
PatronusAI has open-sourced GlobeBench, a benchmark of 1,632 tasks testing whether LLMs can simulate RL environments well enough to train agents on them.
Key result: among 20 models tested, only GPT-6 Astra cleared the bar at 49.6% accuracy — no other model reached 30%, revealing a huge gap in environment-simulation ability.
The poster's sarcastic "sure that's just a coincidence" hints the leader's edge may be no accident.
Related event: PatronusAI Open-Sources GlobeBench for LLM World Modeling(3 posts)→
More from Models
- Steering Layer 18 Head 6 at +1.5 Makes the Model Question If It's Human — Sauers_ · 2026-11-07
- Kelsey Piper: Pangram's loudest critics are writers secretly using AI themselves — TuhinChakr · 2026-11-07
- Autoresearch coding leaderboard: best open model leads but underwhelms, muse open far below promised 1.2 — JoshPurtell · 2026-11-07
- Inkling 2 Small Scores 60% on FrontierMemory, Best Among US Open-Weights Models — ziqiao_ma · 2026-11-07
- Physical Superintelligence claims non-public Claude model proved Aaronson-Ambainis conjecture — Kyrannio · 2026-11-07
- Qwen 4.0 preview reportedly spotted, early tests claim parity with top models — deliprao · 2026-11-07