Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Deep SickThe China-based AI company has released two new reasoning-capable AI models, the V3.2 and V3.2-Speciale, with the latter outperforming the Gemini 3.0 Pro and GPT-5 High in the benchmarks, the company said in a press release on Monday. Currently, version 3.2 is available on the app and web, while version 3.2-Speciale is only accessible via the API.
DeepSeek V3.2 is a follow-up to V3.2-Exp, which stands for beta and was released in September. The Hangzhou-based AI company says version 3.2 outperforms OpenAI’s GPT-5 in inference benchmarks and has “reasonable performance.”On par” with Gemini 3.0 Pro. DeepSeek has not released a test against GPT-5 Pro. That could be because ChatGPT is banned in China. Furthermore, DeepSeek said it is committed to remaining open source, meaning any company can download DeepSeek models for free. But given V3.2 it is lost 685 billion parameterscan only run on a giant server cluster containing millions of dollars’ worth of hardware.
🏆 World-leading thinking
🔹V3.2: Balanced reasoning versus length. Your daily driver with GPT-5 level performance.
🔹 V3.2-Speciale: Maximum inference capabilities. Gemini Rivals-3.0-Pro.
🥇 Gold Medal Performance: V3.2-Speciale achieves Gold level results in IMO, CMO, ICPC World… pic.twitter.com/ZFGzDMshBy– Deep Seek (@deepseek_ai) December 1, 2025
Don’t miss any of our unbiased technical content and lab reviews. Add CNET As Google’s preferred source.
DeepSeek did not immediately respond to a request for comment.
DeepSeek’s release of version 3.2 comes as the company continues to alert US AI companies. The company gained global attention for the first time Earlier this year When I launched a model capable of logical thinking online for free.
It immediately turned the narrative around inference models on its head, showing that it was possible to run smarter, more capable AI models at a fraction of the cost. This spooked Wall Street, with some wondering if OpenAI, Google, and Anthropic weren’t innovating fast enough.
since then, ChatGPT, twin Claude has released Thinking-level models for free, with higher-level “Thinking” models available to paid subscribers.
The company said that DeepSeek-V3.2-Speciale won the gold medal at the 2025 International Mathematical Olympiad (IMO) and International Olympiad in Informatics (IOI).
(Disclosure: Ziff Davis, CNET’s parent company, in April filed a lawsuit against OpenAI, alleging that it infringed Ziff Davis’s copyrights in training and operating its AI systems.)