May 29, 2025:
DeepSeek Unveils Lightweight AI Model for Single GPU - DeepSeek unveiled a compact version of its R1 AI model called DeepSeek-R1-0528-Qwen3-8B, optimized to run on a single GPU. This model, based on Alibaba's Qwen3-8B, shows superior performance compared to similar models, notably outperforming Google's Gemini 2.5 Flash on AIME 2025 math tests and coming close to Microsoft's Phi 4 on HMMT.
Though it lacks the capabilities of larger models, its lower computational requirements make it attractive for various applications. Released under the permissive MIT license, it is accessible for both academic and industrial uses through platforms such as Hugging Face and LM Studio.
