DeepSeek v3.2 closes gap with US frontier models
Subscribe to keep reading.
Trivium Tech keeps you briefed on the latest developments in China tech policy.
Already a subscriber? Log in.
What happened: On December 1, DeepSeek dropped a new model V3.2, pushing closer to US frontier models like OpenAI’s GPT-5 and Google’s Gemini-3.
Per DeepSeek’s technical paper:
“DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-compute variant, DeepSeek-V3.2-Speciale, surpasses GPT-5 an...