SKT has released A.X K2, a large-scale Mixture-of-Experts (MoE) language model. This model boasts 688 billion total parameters with 33 billion active parameters, designed for strong reasoning and instruction-following capabilities while maintaining efficient inference. A key feature is its hybrid reasoning control, allowing users to choose between a 'thinking' mode for complex problem-solving and a 'non-thinking' mode for faster, lower-latency responses. Developed as part of a Korean government initiative, A.X K2 aims to be a frontier-scale model with a deep understanding of Korean language and culture. AI
IMPACT This model's hybrid reasoning control could enable more efficient and adaptable AI applications by allowing users to balance response quality with latency.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Hugging Face Trending Models →
- A.X K2
- Docker Model Runner
- Google Colab
- Kaggle
- mixture of experts
- SGLang
- skt/A.X-K2
- transformers
- vLLM
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
