DeepSeek V3
#6 v Reasoning modelyDeepSeek · v3 · od 26. Dezember 2024 · 55× · naposledy 17. 8. 2026
62
Momentum
DeepSeek-V3 is a Mixture-of-Experts language model developed by DeepSeek AI with 671 billion total parameters, of which 37 billion are activated per token. It was released as an open-source model on December 26, 2024, using Multi-Head Latent Attention (MLA) and a DeepSeekMoE architecture with auxiliary-loss-free load balancing. The model was pre-trained on 14.8 trillion tokens and supports a context window of up to 128,000 tokens; it does not support vision/image input.
Vývoj momenta
19.05.17.08.
Vlastnosti
| Key-Benchmark (%) | MMLU (EM): 88,5% |
| Kontextfenster (Token) | 128.000 Token |
| Lizenz | Code: MIT-Lizenz; Modellnutzung (Base/Chat) unterliegt separater DeepSeek Model License, kommerzielle Nutzung erlaubt |
| Multimodalität | Nur Text, keine Vision-/Bildunterstützung |
| Plattform | API (DeepSeek Platform, OpenAI-kompatibel), Chat unter chat.deepseek.com, Open-Weights auf Hugging Face/GitHub für Self-Hosting |
| Preis pro 1M Token | $0,14 Input / $0,28 Output (Cache-Miss) |
| Release-Datum | 26. Dezember 2024 |