Chinese AI company DeepSeek has unveiled DSpark, an open-source technology designed to speed up large language model responses, VentureBeat reported. DSpark focuses on optimising the inference process without changing the model itself. In DeepSeek tests, perceived speed improved by 60 to 85 percent and total system throughput rose by as much as 661 percent. The technology can also be applied to other open-source models.

A Xiaomi research team has unveiled \'HarnesX,\' a framework that automatically improves AI agent harnesses without changing the underlying model, VentureBeat reported. A harness links large language models to external environments and includes prompts, tools, memory management and execution flow. HarnesX uses a modular design and an AEGIS optimisation engine. In tests across 15 model-benchmark combinations, it improved performance in 14, averaging a 14.5 percent gain.