DeepSeek
DeepSeek: DeepSeek V3
deepseek/deepseek-chat
DeepSeek-V3 is the latest model from the DeepSeek team, building upon the instruction following and coding abilities of the previous versions. Pre-trained on nearly 15 trillion tokens, the reported evaluations reveal that the model outperforms other open-source models and rivals leading closed-source models. For model details, please visit the DeepSeek-V3 repo for more information, or see the launch announcement.
/ 1M tokens
/ 1M tokens
tokens
Limits
Context
163,840
tokens
Max output
16,384
10% of the context window
Parameters
684,531,386,000
Model facts
Author
DeepSeek
Tokenizer
DeepSeek
Modality
Text -> Text
Knowledge cutoff
2024-07-31
Created
Dec 26, 2024
Hugging Face
deepseek-ai/DeepSeek-V3
Publication status
Modalities
Generation controls
Creative tasks
Category performance
1092 Elo
52.7% win
1087 Elo
50.5% win
1090 Elo
50.0% win
1122 Elo
48.5% win
1110 Elo
48.4% win
1055 Elo
43.7% win
976 Elo
38.8% win