Yu Zhang

The DeepSeek-V4.1-Flash model, a multimodal Mixture-of-Experts model with 552B backbone parameters and support for contexts of up to one million tokens, is introduced, substantially improving cost efficiency for agentic workloads and pushing the limits of KV cache compression.

Trie-GRPO, a novel reinforcement learning algorithm based on action prefix trees, which enables step-level advantage estimation, is introduced, which resolves the credit assignment problem by isolating intermediate correct decisions from downstream errors, while effectively balancing exploration efficiency and depth compared to conventional search trees.

SpecFormer transforms recommendation systems by mitigating embedding and attention collapse, leading to superior performance and scalability.

Kimi K3's innovative architecture achieves a 2.5x scaling efficiency improvement, enabling robust performance across diverse long-horizon tasks.

Bridging the reasoning gap, X$^3$-OPD enables audio-language models to outperform their text-based counterparts in logical reasoning tasks.

Player-centric modeling in basketball video analysis reveals that traditional methods miss critical interactions, leading to significant performance improvements with the new PlayNet framework.

LLMs can reason better if you force them to explore *different* ways of being right, not just be more random.