OpenAI's unreleased 'Bel' model reportedly leads internally, but Anthropic wins on shipped models
KanyeWestJS · x · 2026-10-10
A discussion thread clarifies rumors around the math results OpenAI recently published: the model behind them is OpenAI's unreleased internal 'Bel' model, and claims that Anthropic currently has better internal models are false — OpenAI is ahead internally for now, with the two labs expected to keep trading blows on both public and internal models.
Still, the dynamic is a seesaw: on currently released models, most people prefer Anthropic's offerings, and OpenAI's math wins read largely as PR points. If rumors about OpenAI's RSI progress are true, it may be hard to beat, but the poster doubts it's game over.