OpenAI and Anthropic evaluated each others’ models – which ones came out on top

The findings show reasoning models aren’t always more capable than non-reasoning ones, and the biggest safety gaps each company is grappling with.