DeepSeek-V3
by DeepSeek · open-source · Last verified 2026-03-17
DeepSeek's frontier-class MoE model with 671B total parameters and 37B active, trained using FP8 mixed precision for unprecedented cost efficiency. Matches or exceeds GPT-4o and Claude 3.5 Sonnet on key benchmarks.
https://www.deepseek.com ↗B+
B+—Good
Adoption: AQuality: A+Freshness: ACitations: AEngagement: F
Specifications
- License
- MIT
- Pricing
- open-source
- Capabilities
- text-generation, code-generation, reasoning, math, multilingual, function-calling
- Integrations
- huggingface, vllm, ollama, langchain, llama-index
- Use Cases
- frontier-research, code-generation, complex-reasoning, enterprise-deployment, math-problem-solving
- API Available
- Yes
- Parameters
- 671B (37B active)
- Context Window
- 128K tokens
- Modalities
- text
- Training Cutoff
- Late 2024
- Tags
- llm, open-source, moe, frontier, fp8-training, deepseek
- Added
- 2026-03-17
- Completeness
- 100%
Index Score
72.8Adoption
82
Quality
90
Freshness
85
Citations
88
Engagement
0