Computing power surges 14 times

The computing power surges 14 times! OpenAI officially releases the GPT-5.6 Sol “ultra-fast” mode OpenAI recently officially launched the highly anticipated GPT-5.6Sol “ultra-fast” mode. Prior to this, this version was previewed internally for some specific customers in July, and now the output speed of this mode can reach up to 750 tokens per second. The overall processing speed is up to 14 times faster than the standard mode without changing to a smaller or lower-performance model. This technological breakthrough is mainly due to Cerebras’ wafer-level engine architecture support. This architecture has a built-in 44GB of on-chip static random access memory, which successfully breaks the common data transmission bottleneck in previous GPU-based systems. According to relevant test data, its running speed not only greatly exceeds that of similar competing products, but also processed all 2,500 questions in the "Ultimate Human Exam" in only about 11 hours, demonstrating extremely high computing efficiency. This new model is now ideally suited for high-intensity workflows such as incident response, financial research, real-time customer support, programming, and complex research. Since computing power resources are still limited, OpenAI will gradually evaluate and issue access rights based on the customer's workload suitability and computing power availability. via AI News (author: AI Base)