DeepSeek-V3-0324

#138 en Modèles de langage open source

DeepSeek · v3 · 0324 · 9× · vu le 30 juin 2026

0

Momentum

DeepSeek-V3-0324 is an open-weight Mixture-of-Experts language model released by DeepSeek-AI with 671 billion total parameters, of which 37 billion are active per Token. It is an incremental update of DeepSeek-V3 with identical model architecture but improved Benchmark results in mathematics, code, and Reasoning, along with a refined writing style. The model was released on March 24, 2025, with weights openly available under the MIT license on Hugging Face, and is accessible via the DeepSeek API as well as numerous third-party providers. It processes text only, with no image or audio inputs.

Preuves (9)