- Chinese startup DeepSeek launched two powerful new AI models that rival GPT-5 and Gemini 3 Pro
- The models are open source, inexpensive to run, and designed for real-world reasoning using built-in tools.
- Its launch challenges US technological dominance and reignites global AI competition
Chinese startup DeepSeek has casually launched two massive new AI models into the already simmering international arms race: DeepSeek-V3.2, a model built for everyday reasoning, and DeepSeek-V3.2-Speciale, a high-octane variant that reportedly outperformed top American models in elite math and coding competitions. Not only that, but they released them under an open source license.
What makes this movement notable is not only the models and their capabilities, but also how they were released. US companies like OpenAI and Google rely on powerful, and often expensive, models that rely on private APIs and red team testing for the most advanced models. DeepSeek has weaponized openness.
DeepSeek-V3.2 reportedly matches or surpasses GPT-5 and Gemini 3 Pro in extensive reasoning, tool usage, and dense problem solving, including competitions such as the International Mathematics Olympiad and the ICPC World Finals. The “Speciale” version scored 99.2% in the Harvard-MIT Mathematics Tournament, 73% in software bug fixes, and posted gold medal results in multiple international benchmark tests even without Internet access or external tools.
The trick behind this performance is a clever architectural trick called DeepSeek Sparse Attention, or DSA. Traditional transformer models become computationally bloated as context length increases, and have to consider each word in a document in relation to all the others. DSA reduces costs by focusing only on the most relevant parts of the entry, essentially skimming rather than reading every word. That alone reduces long document costs by up to 70%, making the model relatively cheap.
🚀 DeepSeek-V3.2 and DeepSeek-V3.2-Speciale released: reasoning models built for agents! 🔹 DeepSeek-V3.2: Official successor to V3.2-Exp. Now available in apps, web and API.🔹 DeepSeek-V3.2-Speciale: Pushing the limits of reasoning capabilities. Only API for now.📄 Technology… pic.twitter.com/SC49UdmCZvDecember 1, 2025
This is important for real people because cost is everything in AI implementation. Most current border models are trapped behind payment barriers and restricted access. But DeepSeek’s latest models and their 128,000-token contextual windows can be downloaded and modified for free. A solo developer or a team of students can tinker with systems that just a few months ago would have required a lab and a huge cloud budget.
DeepSeek’s advancement in “tool thinking” is especially notable. Most AI agents struggle to juggle multiple tools because each action resets their internal reasoning. DeepSeek solved this problem by preserving memory across all tools. The company trained the model using more than 85,000 complex synthetic instructions to work with tools such as real web browsers and coding environments.
That’s a level of readiness for real-world tasks that most current chatbots simply aren’t designed for. It’s one thing to summarize a recipe. It’s another thing to plan a multi-day vacation on a strict budget, with interdependent accommodation and food restrictions, and do it while testing code snippets and checking exchange rates.
International AI Rivalry
License setup could be even more problematic. By using the MIT open source license, DeepSeek has made it legally possible for anyone to copy, remix or commercialize its models. This goes directly against the current trend of protecting model weights as intellectual property, citing security, risk of misuse and corporate secrecy.
However, openness does not mean transparency. That’s why German regulators have tried to block DeepSeek over concerns about data transfer. Italy banned the app earlier this year and US lawmakers want it out of government agencies entirely. DeepSeek is a Chinese company and the geopolitical context is of great importance. But let’s assume that DeepSeek’s models really do offer cutting-edge performance at a fraction of the cost, and you don’t care about the geopolitical baggage. What exactly do American companies offer that are worth the margin?
For now, the Speciale variant of DeepSeek is only available through a temporary API. But in mid-December, it will be merged into the broader version V3.2 and will be accessible to everyone. If the last few years were defined by ChatGPT’s friendly introduction to AI, this release seems a stark reminder: the gloves are off and the global AI race is no longer just about features, but about access, cost and control.
Follow TechRadar on Google News and add us as a preferred source to receive news, reviews and opinions from our experts in your feeds. Be sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form and receive regular updates from us on WhatsApp also.
The best business laptops for every budget
