DeepSeek-V3 at a glance
DeepSeek-V3 was a mixture-of-experts model with 37B of 671B parameters active, trained on 14.8T tokens in 2.788M H800 GPU hours, with open weights that allowed commercial use.
What changed with DeepSeek-V3
DeepSeek released DeepSeek-V3 on December 26, 2024. It was a mixture-of-experts model with 671B total parameters and 37B active per token, trained on 14.8 trillion tokens. DeepSeek said it generated about 60 tokens per second, three times faster than V2.
The model card stated that full training took only 2.788 million H800 GPU hours, a figure that drew wide attention to training efficiency. Weights went to Hugging Face and code to GitHub under the MIT license, with the model under DeepSeek's model license, which allows commercial use. API prices stayed at V2 levels until February 8, 2025.
DeepSeek-V3 in numbers
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Total / active parametersas of · Source for Total / active parameters: api-docs.deepseek.com (opens in a new tab)
- 671B total, 37B active
Where to compare DeepSeek-V3
Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.
General leaderboards
Why DeepSeek-V3 matters for businesses
For businesses, V3 reset expectations for what strong models should cost to train and run, and set up the R1 reasoning model a month later. Companies gained a powerful open option but had to weigh data-residency and vendor-risk questions before using DeepSeek's own hosted API.
Sources for this story
- Introducing DeepSeek-V3 (opens in a new tab)Official
DeepSeek API Docs · api-docs.deepseek.com
- DeepSeek-V3 model card (opens in a new tab)Official
DeepSeek on Hugging Face · huggingface.co
- DeepSeek-V3 repository (opens in a new tab)Official
DeepSeek on GitHub · github.com
- DeepSeek V3 (Dec) model page (opens in a new tab)Benchmark
Artificial Analysis · artificialanalysis.ai
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #llm
- #deepseek
- #open-weights
- #mixture-of-experts