DeepSeek Unveils V3.2 AI Models Matching OpenAI, Google

Chinese artificial intelligence startup DeepSeek unveiled two advanced versions of its large language model on Monday, claiming performance on par with leading systems from OpenAI and Google. The Hangzhou-based company released DeepSeek V3.2 and DeepSeek V3.2 Speciale, building on an experimental iteration introduced in September.

The updates enhance the model's ability to combine reasoning with autonomous actions, according to DeepSeek. The base model, V3.2, matches OpenAI's GPT-5 across multiple reasoning benchmarks, the company stated in a technical report titled "DeepSeek V3.2: Pushing the Frontier of Open Large Language Models." It also performs similarly to Moonshot AI's Kimi K2 Thinking on those metrics. DeepSeek described V3.2 as its first to integrate thinking directly into tool use, enabling interactions with search engines, calculators and code executors in both thinking and non-thinking modes.

DeepSeek V3.2 Speciale targets mathematical computations and extended reasoning tasks. The company said it equals Google's Gemini 3 Pro and achieved gold medal-level results on standardized tests, including the International Mathematical Olympiad and the International Olympiad...