Upstage has released Solar Open 2, a 250 billion parameter open-weight large language model designed for agentic tasks. The model features a Hybrid-Attention Mixture-of-Experts (MoE) architecture, enabling efficient inference with a 15 billion parameter activation per token and supporting a 1 million token context window. Solar Open 2 is optimized for productivity, document analysis, and coding, with multilingual capabilities in English, Korean, and Japanese. AI
IMPACT This release offers a powerful, efficient open-weight model for agentic tasks, potentially accelerating development in areas like productivity and coding.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=2 ai=1.0]
- Docker
- Google Colab
- Hugging Face
- Kaggle
- OpenAI
- SGLang
- Solar Open 2
- transformers
- Upstage
- upstage/Solar-Open2-250B
- vLLM
- Solar Open 1
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →