DeepSeek-V3
High-performance open-weight MoE LLM with 671B total parameters.
About
DeepSeek-V3 is an open-weight Mixture-of-Experts language model with 671 billion total parameters and 37 billion active per token. It uses Multi-head Latent Attention and the DeepSeekMoE architecture with an auxiliary-loss-free load-balancing strategy and a multi-token prediction objective, and trains efficiently with FP8 mixed precision. It is competitive with leading proprietary models on benchmarks. Released under the DeepSeek license.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Category
- Large Language Models (LLMs)
- Price
- Free
- Platform
- Local/Desktop
- Difficulty
- Advanced (4/5)
- License
- DeepSeek License
- Minimum VRAM
- 24 GB
- Added
- Apr 3, 2026
Related Tools
Open-source code LLM family by IBM for enterprise code generation.
Hybrid SSM-Transformer model by AI21 Labs combining Mamba with attention layers.
State-space model for language with linear-time sequence modeling.
Family of small language models by Hugging Face for on-device use.
Open-source multimodal LLM competitive with commercial models by Shanghai AI Lab.
Open-weight LLM by Meta in 8B and 70B sizes with strong general capabilities.