News

It achieved an 8.0% higher win rate over DeepSeek R1, suggesting that its strengths generalize beyond just logic or math-heavy challenges.