DeepSeek is back! Chinese startup launches two new models to take on Gemini

DeepSeek, a prominent AI startup from China, has introduced two innovative models aimed at rivaling the established dominance of Gemini and ChatGPT in the chatbot arena. Earlier this year, the company gained significant traction with its R1 and V3 models, which demonstrated competitive performance against leading Western counterparts while being developed at a substantially lower cost. However, in recent months, DeepSeek has faced challenges in matching the capabilities of its competitors, particularly as Google's Gemini and Nano series have excelled in handling text, image, and video tasks. OpenAI has also escalated the competition with the release of several advanced models, including the latest GPT 5.1. Other Chinese firms, such as Qwen and Kimi, have also made headlines with their impressive AI offerings. The landscape may shift with the launch of the DeepSeek V3.2 and V3.2 Speciale models. These new iterations are designed as reasoning-first models, tailored for agent applications. DeepSeek claims that the V3.2 series can achieve performance levels comparable to GPT 5 in various general tasks. Furthermore, the V3.2 Speciale model reportedly showcases reasoning capabilities that rival Google's Gemini 3 Pro, particularly in complex problem-solving situations. Remarkably, the V3.2 Speciale model earned top honors at the 2025 International Mathematical Olympiad and the International Olympiad in Informatics—accolades previously dominated by proprietary models from Google and OpenAI. In addition, it outperformed both GPT 5 High and Gemini 3 Pro in rigorous math assessments like AIME and HMMT, and excelled in coding logic challenges on platforms like CodeForces, where it surpassed all competing AI models. In the challenging Humanity's Last Exam benchmark, the DeepSeek V3.2 Speciale achieved a score higher than GPT 5 High but slightly lagged behind Gemini 3 Pro. The V3.2 model has now taken over from the V3.2 Experimental version, becoming the default option on the chatbot's website and application. Users eager to explore the new capabilities can access the V3.2 model by logging into DeepSeek's platform. Meanwhile, the V3.2 Speciale is currently available for API use only, with no announced timeline for broader user access.