Discussion on model exploratory behavior and pass@k metrics

scaling01 · x · 2026-08-21

Discussion on model performance regarding pass@k metrics, noting that OpenAI was previously seen as 'exploration maxxing' while Anthropic was 'exploitation maxxing,' but recent Anthropic models have become more exploratory. The author suggests this exploration maxxing is mostly RL-related, but pre-training builds the base for pass@k performance.

Related event: Debate Over pass@k: OpenAI's Exploration vs Anthropic's Exploitation(2 posts)→

Original post →

More from Models

Models channel →