DeepSeek unveils 2 new models, claims performance on par with GPT-5, Gemini

The latest launch comes as the Hangzhou-based company didn't deliver any powerful models while its rivals, Google's Gemini and OpenAI's ChatGPT, gained a lot of traction with their latest models

Don't want to miss the best from Business Standard?

China's artificial intelligence (AI) startup DeepSeek on Monday launched two new models, DeepSeek-V3.2 and DeepSeek-V3.2 Speciale, designed to combine high efficiency with advanced reasoning and tool-use capabilities.

 

The latest launch comes as the Hangzhou-based company didn't deliver any powerful models recently while its rivals, Google's Gemini and OpenAI's ChatGPT, gained a lot of traction with their latest models -- GPT-5 and Gemini 3.0-Pro. Meanwhile, in China, AI firms Qwen and Kimi dominated the race.

Enhanced reasoning capabilities

In an X post on Monday, DeepSeek said, "Launching DeepSeek-V3.2 & DeepSeek-V3.2-Speciale — Reasoning-first models built for agents!"

 

DeepSeek-V3.2 is built on the same usage pattern as that of the company's experimental model V3.2-Exp, which was launched a few weeks ago. DeepSeek-V3.2 Speciale is aimed at further pushing the boundaries of reasoning capabilities.

 

According to the company, DeepSeek-V3.2 is their first model that builds reasoning directly into tool-use and works in both thinking and non-thinking modes. The model can also generate large-scale agent training data, spanning more than 1,800 environments and over 85,000 complex instructions. 

New models rival GPT-5, Gemini 3.0-Pro

The models use DeepSeek Sparse Attention (DSA), a faster attention mechanism that reduces computing costs while maintaining strong performance, especially for long inputs.

 

The company further said that its V3.2 model, with its balanced inference, matches GPT-5 performance. Meanwhile, V3.2-Speciale, with its maxed-out reasoning capabilities, rivals Gemini-3.0-Pro. This version achieved gold-medal level performance in the 2025 International Mathematical Olympiad and International Olympiad in Informatics.

DeepSeek's meteoric rise

Earlier this year, DeepSeek shook the US tech giants with its rapid rise. In January, news of the DeepSeek chatbot’s top-tier performance wiped hundreds of billions of dollars off major technology stocks.

 

A key factor behind its success was low cost. DeepSeek matched the capabilities of its rivals at only a fraction of the price. At that time, however, it had not yet reached artificial general intelligence (AGI), the level at which AI can reason like humans, a threshold achieved by OpenAI and other leading AI companies. Now, it appears that DeepSeek has closed that gap.