December began with a big announcement by DeepSeek. The Chinese AI startup had recently released an experimental third version of their AI model on September this year, only to release the official third edition of DeepSeek AI this week. The company took to their official social media platform to share the news, informing the new capabilities the two AI models are prepared to execute.
With this, another big wave of innovations seems to have swept across the AI tech industry as DeepSeek came up with a dedicated model for mathematical components, and another with an equally crucial development. This could certainly pit their latest releases againstOpenAI’s GPT-5.1 and Google’s Gemini-3 Pro that were released last month.
DeepSeek announces two new AI models to rival Gemini and ChatGPT
On December 1, the DeepSeek team took to their official X account to share an important update. They announced the release of their two new AI models – DeepSeek-V3.2 and DeepSeek-V3.2-Speciale.
The former model is the direct successor of the former DeepSeek-V3.2-Exp. Released in September this year, the latter was an experimental project launched to evaluate its performance to prepare the final product.
🚀 Launching DeepSeek-V3.2 & DeepSeek-V3.2-Speciale — Reasoning-first models built for agents!
— DeepSeek (@deepseek_ai) December 1, 2025
🔹 DeepSeek-V3.2: Official successor to V3.2-Exp. Now live on App, Web & API.
🔹 DeepSeek-V3.2-Speciale: Pushing the boundaries of reasoning capabilities. API-only for now.
📄 Tech… pic.twitter.com/SC49UdmCZv
DeepSeek-V3.2 is meant to imitate human-like reasoning while employing tools like the code executors, search engines and calculators to deliver quality results in response to prompts of any kind. Programmed on the basis of Reinforced Learning (RL), one of DeepSeek’s primary goals is to develop an LLM that would learn to improve itself on the basis of former inputs received and outputs provided by the same.
As per a report by Bloomberg, “The startup says the new service matches the performance of OpenAI Inc.’s flagship GPT-5 across multiple reasoning benchmarks, suggesting China’s open-source systems remain competitive with Silicon Valley’s frontier proprietary models on at least certain metrics.”
Simultaneously, this also throws light on the accessbility of information across the industry for AI startups like DeepSeek and, very recently, Sapient Intelligence to contribute positively to the ongoing AI revolution.
Their second product, DeepSeek-V3.2-Speciale, is more oriented towards mathematical components as well as focusing on AI capabilities for long-thinking processes. Although it is only available in API mode, the V3.2-Speciale has reportedly come close to Google’s Gemini 3 Pro, as per Bloomberg’s report. The new AI model was also tested to take part in standardized mathematical and reasoning tests.
Related: Gen Z Founders of Sapient Inc. Turn Down Elon Musk’s Multimillion-Dollar Offer
DeepSeek continues to challenge western innovations in AI tech
The new releases by DeepSeek this month hints at a technical turn towards investing more money and intellect in developing the AI models to be more efficient. This efficiency refers to the AI’s capability of processing more data and preparing responses in a more comprehensive and exhaustive manner while also fulfilling the user requirements conveyed through the prompts.
This marks yet another moment this year when DeepSeek made it big in the AI tech industry. Although most companies in the Silicon Valley and beyond in the West are seemingly engrossed on more data and more resources, the Chinese AI startups appear to be focused on the quality and manner(s) in which the AI models will perform when presented with varying kinds of prompts.
The decisive factor in this context is the employment of the RL methodology, which has become a crucial factor in development of AI. After all, an LLM cannot merely be fed more information unless it is capable of sifting through it and picking up the exact pieces necessary as per the prompt with the help of reasoning.
Also Read: OpenAI CEO Sam Altman Plans To Make A “Very Aggressive” Bet
