Cursor's New Model Achieves Pareto Dominance After Three Weeks on Colossus

What was said

"Cursor's Composer 2.5 model came out this week. This is Pareto dominant. This is just 3, 4 weeks of doing reinforcement learning on Colossus-2 with Cursor's data. This is 3 or 4 weeks and it is Pareto dominant."

Cursor's Composer 2.5 model achieved Pareto dominance over all competing coding models after only 3-4 weeks of reinforcement learning on SpaceX's Colossus-2 compute cluster. Gavin Baker emphasized that Cursor allegedly possesses more proprietary tokens of coding data than exist on the public internet, and the dramatic improvement demonstrates the power of combining massive compute with proprietary training data.