VIDRAFT has released POCKET-Darwin-180B, a highly compressed version of their Darwin-180B-RSI model, which now runs on consumer hardware like laptops and mini PCs without requiring a dedicated GPU. This 111 GB model, a significant reduction from its original 360 GB size, maintains top performance across seven Hugging Face leaderboards. The model's efficiency is attributed to its Mixture-of-Experts architecture and memory-mapping techniques, making it accessible for organizations with strict data privacy requirements. AI
IMPACT Enables deployment of top-tier LLMs on air-gapped systems and consumer devices, bypassing the need for expensive GPU infrastructure.
RANK_REASON Release of a compressed frontier-class model by VIDRAFT that achieves top leaderboard performance and runs on consumer hardware. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Alibaba Group
- Darwin-180B-RSI
- Hugging Face
- llama.cpp
- ModelScope
- NVIDIA H100
- OpenBMB
- POCKET-Darwin-180B
- Qwen3.8-Flash-Next
- RTX 5060 Laptop
- Shanghai Artificial Intelligence Laboratory
- VIDRAFT
- Zhipu AI
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →