Story perspectives

DeepSeek Launches Powerful AI Models Amid Regulatory Challenges

12/3/2025

Summary

Story summary

DeepSeek launched two models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, claiming they match or exceed GPT-5 and Gemini-3.0-Pro.
The DeepSeek-V3.2 model serves as a reasoning assistant.
The DeepSeek-V3.2-Speciale variant won gold in the International Mathematical Olympiad.
DeepSeek introduced a Sparse Attention mechanism that reduces computational costs.
Regulatory challenges in Europe and the United States may hinder global adoption, and DeepSeek's open-source approach could disrupt the AI market.

1 of 1

Story summary

DeepSeek launched two models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, claiming they match or exceed GPT-5 and Gemini-3.0-Pro.
The DeepSeek-V3.2 model serves as a reasoning assistant.
The DeepSeek-V3.2-Speciale variant won gold in the International Mathematical Olympiad.
DeepSeek introduced a Sparse Attention mechanism that reduces computational costs.
Regulatory challenges in Europe and the United States may hinder global adoption, and DeepSeek's open-source approach could disrupt the AI market.