DeepSeek details DSpark, a speculative decoding framework for its V4 models, saying it speeds up AI inference by up to 85% and was tested on Gemma and Qwen (Ben Jiang/South China Morning Post)

DeepSeek has introduced DSpark, a speculative decoding framework that accelerates AI inference speeds by up to 85% across multiple model architectures including its V4, Gemma, and Qwen models. This breakthrough directly impacts enterprise AI deployment costs and latency, enabling organizations to deliver faster AI-powered applications while reducing computational infrastructure requirements. For IT leaders, this represents a significant opportunity to optimize AI workload performance and total cost of ownership across existing and new generative AI initiatives.