Poolside has released Laguna XS 2.1, a 33B parameter Mixture-of-Experts model optimized for coding and long-horizon tasks on local machines. This updated model features a 3:1 ratio of Sliding Window Attention to global attention layers, FP8 quantization for its KV cache to reduce memory usage, and native reasoning support for interleaved tool calls. With only 3B activated parameters per token, Laguna XS 2.1 is designed to run on systems with as little as 36GB of RAM and is compatible with various libraries and inference providers. AI
IMPACT This model's focus on local execution and agentic capabilities could accelerate adoption of advanced AI for coding and complex tasks on personal hardware.
RANK_REASON Model release from a known AI lab (Poolside). [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Hugging Face Trending Models →
- Docker
- Google Colab
- Hugging Face
- Kaggle
- OpenAI
- OpenRouter
- Poolside Laguna XS 2.1
- SGLang
- transformers
- vLLM
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
