#128K context

'Meituan released LongCat Flash Omni, a 560B open-source omni-modal model activating ~27B parameters per token that supports real-time audio-visual interaction and long 128K context understanding.'

'Alibaba open-sourced Tongyi DeepResearch-30B-A3B, a 30B-parameter MoE LLM tailored for long-horizon, tool-augmented web research with 128K context and strong benchmark scores.'

'NVIDIA's Nemotron Nano 2 delivers hybrid Mamba-Transformer LLMs that run up to 6× faster and support 128K-token context on a single A10G GPU, with most training data and recipes open-sourced.'