DeepSeek — Model Parameters: 671 billion
The source text directly confirms both key claims: (1) DeepSeek-V3 has 671B total parameters, and (2) 37B parameters are activated per token. The abstract states this explicitly, and the introduction reiterates 'a large Mixture-of-Experts (MoE) model with 671B parameters, of which 37B are activated for each token.' The exact stored value of 671,000,000,000 is numerically equivalent to 671B. The source is the official DeepSeek-V3 Technical Report (arxiv 2412.19437) released in December 2024, matching the 'as of 2024-12' temporal specification in the claim.
Our claim
entire record- Subject
- DeepSeek
- Property
- Model Parameters
- Value
- 671 billion
- As Of
- December 2024
- Notes
- DeepSeek-V3: 671B total parameters (MoE architecture), 37B activated per token. Released December 2024.
Source evidence
1 src · 1 checkNoteThe source text directly confirms both key claims: (1) DeepSeek-V3 has 671B total parameters, and (2) 37B parameters are activated per token. The abstract states this explicitly, and the introduction reiterates 'a large Mixture-of-Experts (MoE) model with 671B parameters, of which 37B are activated for each token.' The exact stored value of 671,000,000,000 is numerically equivalent to 671B. The source is the official DeepSeek-V3 Technical Report (arxiv 2412.19437) released in December 2024, matching the 'as of 2024-12' temporal specification in the claim.