Cursor announces the launch of a brand-new large model with over 1.5 trillion parameters. Cursor's CEO confirmed that this model was pre-trained on a compute cluster of over 100,000 GPUs, with its performance and parameter scale fully aligned with Anthropic's Opus core series and OpenAI's flagship GPT series. Leaked supply chain data reveals that the current market-hyped closed-source frontier models, including GPT-5 to GPT-5.5 and Claude Opus 4.5 to 4.8, all have actual parameter scales below 2 trillion (2T). This means that the performance ceiling of this generation of closed-source flagships can technically be matched by open-source or hybrid expert (MoE) architectures of similar scale, such as DeepSeek-V4-Pro. The long-term technical barriers and high API premiums previously built by major players are being rapidly eroded by the high-concurrency, compute-intensive nature of open-source power.