Model Benchmarks

Compare the performance of different AI models across standardized benchmarks. Higher scores generally indicate better performance, but context matters for each benchmark.

Benchmarks

ScienceWorld Goal Achievement

GPT-J-6B achieved a 3.5x improvement over reinforcement learning-based approaches in planning for simulated robots to achieve 30 classes of goals in ScienceWorld, a text game simulator for elementary science experiments.

Rank	Model	Provider	Score (times improvement over RL)	Date	Source
1	GPT-J-6B	EleutherAI	3.50	May 4, 2025	View