Records found: 6
'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.'
'Ling 2.0 is a reasoning-first sparse MoE family from Ant Group that keeps per-token compute low with a 1/32 activation recipe while scaling from 16B to 1T parameters.'
'MiniMax-M2 is a compact MoE released with open weights on Hugging Face, optimized for coding and agentic tool use with about 10B active parameters per token and claims of ~2x speed at ~8% of Claude Sonnet cost.'
A detailed technical comparison of Alibaba's Qwen3 30B-A3B and OpenAI's GPT-OSS 20B MoE transformer models, highlighting architectural differences and use case recommendations.
FlexOlmo introduces a modular framework that allows training large language models on private datasets without data sharing, achieving strong performance while respecting data governance and privacy constraints.
Huawei has introduced Pangu Ultra MoE, a 718 billion parameter sparse language model optimized for Ascend NPUs using simulation-driven architecture and advanced system-level optimizations to achieve high efficiency and performance.