How DeepSeek’s DSpark Makes LLMs 85% Faster Without Retraining

DeepSeek has introduced DSpark, an open-source speculative decoding method that accelerates LLM inference by up to 85% ...
i-SCOOP

LongCat 2.0 by Meituan pushes trillion-parameter models onto Chinese chips

LongCat 2.0 is Meituan's 1.6T-parameter MoE model with a native 1M-token context, trained entirely on domestic AI ASIC ...

Explained: How China is narrowing the AI gap with the US one model at a time

Just when the AI industry’s attention seemed fixed on OpenAI, Google and Anthropic, a new Chinese model has stolen the spotlight. The arrival of GLM 5.2 underscores the rapid pace of AI innovation in ...