DeepSeek releases new AI systems challenging industry leaders
DeepSeek, a technology company based in Hangzhou, has introduced two new versions of its artificial intelligence platform, underscoring China’s growing capacity in open-source development and advanced machine learning, reports a Qazinform News Agency correspondent.
The company announced the release of DeepSeek V3.2 and V3.2 Speciale, both designed to compete with leading systems developed by Google and OpenAI. According to the developers, V3.2 integrates autonomous tool use with enhanced reasoning, enabling the model to work with search engines, code execution tools and calculators without human intervention. On several logic and reasoning benchmarks, it performs at a level comparable to GPT-5.
The Speciale version targets mathematical and analytical tasks. DeepSeek reports that it demonstrates results similar to Google’s Gemini-3 Pro and performs competitively on tests associated with the International Mathematical Olympiad. The company notes that both versions support flexible modes of operation, combining structured reasoning with rapid problem solving.
DeepSeek earlier attracted wide attention after the launch of its R1 model, which highlighted the potential of efficient training methods in the open-source ecosystem. The firm states that the V3 line was trained at a substantially lower cost than comparable Western models, supported by new architectural features that improve work with long documents and reduce operating expenses.
The V3.2 model is already available through applications, the web platform and API, while access to the Speciale version is currently provided through API services.
Earlier, Qazinform News Agency reported that OpenAI expects around 8.5% of an estimated 2.6 billion weekly users, or about 220 million people, to subscribe to its chatbot by 2030, placing ChatGPT among the largest subscription services worldwide.