Sources
Loading...
Additional media
Loading...
DeepNewz, mobile.
People-sourced. AI-powered. Unbiased News.
China has unveiled the Hunyuan-TurboS, the first ultra-large Hybrid-Transformer-Mamba MoE model, which reportedly outperforms established competitors such as GPT-4o-0806 and DeepSeek-V3 in areas of math, reasoning, and alignment. This new model addresses challenges faced by traditional pure Transformer models, particularly in long-text training and inference due to their O(N²) complexity and KV-Cache issues. The model's introduction follows Alibaba's QwQ-32B, which has also demonstrated superior performance in math and reasoning tasks compared to larger rivals.
HOLY MOLY, China is cooking! Hunyuan-TurboS – the first ultra-large Hybrid-Transformer-Mamba MoE model! (what a name lol!) - Outperforms GPT-4o-0806, DeepSeek-V3, and open-source models on Math, Reasoning, and Alignment Come on now OpenAI, it's time for the next big thing!Now… https://t.co/eb37CZIP45 https://t.co/03iALVQnh5
🚨 China just dropped another model… I repeat, another model! 🔥🔥 Hunyuan-TurboS – first ultra-large Hybrid-Transformer-Mamba MoE model. - outperforms GPT-4o-0806 and DeepSeek-V3. https://t.co/54IxteWeG8 https://t.co/M80jU7zwoq
Hunyuan releases TurboS – the first "ultra-large Hybrid-Transformer-Mamba MoE model". https://t.co/csYzzOz548
