DeepSeek challenges Google and OpenAI with new AI model

(CLO) DeepSeek has just announced its most powerful model variant, DeepSeek-V3.2-Speciale, which is said to be on par with Google DeepMind's new Gemini 3 Pro model in some tasks.

In an announcement Monday, Hangzhou-based DeepSeek said V3.2-Speciale is on par with Google's Gemini 3 Pro (launched two weeks ago) in terms of inference capabilities.

DeepSeek trains AI models at surprisingly low cost.

Meanwhile, the V3.2-Speciale baseline model — which was released on the same day as V3.2 — performed on par with OpenAI's GPT-5 (released in August).

DeepSeek claims V3.2-Speciale achieved a gold medal in the International Mathematical Olympiad, an achievement previously only achieved by unreleased internal models from OpenAI and Google DeepMind.

DeepSeek has released the open source code for the V3.2 model on its Hugging Face developer platform. However, according to the announcement, V3.2-Speciale is only accessible via an application programming interface (API).

DeepSeek achieved this feat at a time when the Chinese company had limited access to advanced semiconductor chips.

Source: https://congluan.vn/deepseek-thach-thuc-google-va-openai-voi-moi-hinh-ai-10320085.html