China-based DeepSeek is making waves in the AI world with the release of two new models. These are DeepSeek-V3.2 and DeepSeekMath-V2; which the company claims can match or outperform leading proprietary systems from Google and OpenAI, all while offering a more cost-effective alternative that could reshape how AI is developed and used.
DeepSeek unveiled the latest version simply called DeepSeek-V3.2, and the startup says this new service matches the performance of OpenAI’s flagship GPT-5 across multiple reasoning benchmarks . What makes this interesting is that DeepSeek managed to achieve these results while spending far less money than the American companies that dominate the AI space.
The company is based in Hangzhou, China, and is owned by a hedge fund called High-Flyer. When DeepSeek first launched its R1 model in January, it caught everyone by surprise. Nobody expected a relatively unknown Chinese startup to create something that could compete with OpenAI and Google. But the numbers speak for themselves. DeepSeek claims that it trained its V3 model for around six million dollars, far less than the one hundred million dollar cost for OpenAI’s GPT-4 , and using much less computing power than what Meta needed for its similar model.
The newest version that just came out includes something called DeepSeek Sparse Attention, which is a fancy way of saying the AI is now better at handling long documents and conversations. This feature cuts the cost of running the AI in half compared to the previous version . For people using the AI, this means faster responses and lower bills when accessing the service.
What really has people talking is how DeepSeek is changing the game when it comes to making AI more accessible. The company releases all its models as open source, which means anyone can look at how they work, learn from them, and build their own improvements. This is very different from companies like OpenAI and Google, which keep their models locked away as trade secrets.
When DeepSeek’s chatbot launched in January, it became so popular that it actually surpassed ChatGPT as the most downloaded free app on Apple’s App Store in the United States. This success sent shockwaves through the tech industry. Nvidia, the company that makes the expensive chips used to power AI systems, saw its stock price drop sharply because investors worried that if AI can be built this cheaply, maybe companies won’t need to buy as many expensive chips.
The company is not stopping with just these models either. DeepSeek is working on even more advanced features that will let the AI carry out tasks on behalf of users with very little guidance. Think of it like having a smart assistant that can figure out what you need and take multiple steps to get it done, learning and improving as it goes along.
For businesses and developers, DeepSeek’s approach is particularly appealing. The lower costs mean smaller companies can now afford to experiment with AI and build new applications without needing massive budgets. This levels the playing field and could lead to a wave of innovation from places that were previously locked out of the AI revolution because it was too expensive.
The technology also works well with Chinese-made chips, which is important because the United States has placed restrictions on what kinds of advanced chips can be sold to China. By making AI that runs efficiently on less powerful hardware, DeepSeek has found a way around these limitations while also proving that you don’t always need the most expensive equipment to create powerful AI systems.
Some countries have raised concerns about DeepSeek and banned government agencies from using it, citing national security worries. Italy, the United States, and South Korea have all taken this step. Despite these concerns, the technology continues to gain users around the world who are attracted by its performance and low cost.
The impact of DeepSeek goes beyond just providing another AI chatbot option. The company is showing that artificial intelligence doesn’t have to be controlled by a handful of American tech giants. By sharing their work openly and proving that powerful AI can be built affordably, they’re changing the conversation about who gets to participate in the AI revolution.
DeepSeek’s approach focusses on efficiency and open-source accessibility. Their models are recognised for being built for a tenth of the cost of other prominent models and being distributed under an MIT license, which allows them to be freely used and customised commercially. This method seeks to democratise access to powerful AI capabilities, posing a significant threat to the proprietary models given by large IT companies.
Researchers and developers can examine the models on Hugging Face and use them through the DeepSeek API.
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
