Up to 750 Tokens Per Second: OpenAI Launches Ultrafast Mode, GPT-5.6 Sol AI Speeds Up 14x

OpenAI has officially introduced an Ultrafast mode for its flagship model, GPT-5.6 Sol. This mode increases standard processing speed by 14 times, with peak output reaching up to 750 tokens per second, significantly optimizing the real-time response performance of large models.

Behind this performance breakthrough is the computing power support of Cerebras. The intervention of dedicated AI chips has resolved the bottleneck of high-concurrency inference under traditional architectures, enabling ultra-large parameter models to handle denser computing tasks while maintaining a high level of intelligence.

In the past, users often had to compromise between model capability and response speed, choosing smaller models to gain speed. The Ultrafast mode breaks this limitation, allowing developers to achieve smooth interaction without sacrificing model quality, thereby improving the processing efficiency of complex tasks.

This update will further promote the application of AI in real-time dialogue, code generation, and automated workflows. Faster inference speed means lower latency, which helps enhance user trust and reliance on generative AI tools, accelerating technology adoption.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.