OpenAI Cries Foul as DeepSeek Pushes Open Source AI Forward

Imagine raising and spending billions only to have someone’s side project beat your performance with a fraction of the budget. That’s the story brewing around DeepSeek, an AI model built in China that might give OpenAI and Claude a run for their money. Trained on a relatively lean $5.5 million (nearly nothing in the AI world), DeepSeek burst onto the scene with chain-of-thought reasoning and an open-source approach. Even more outrageous, it’s super cheap.

IP theft or sour grapes?

Not everyone is celebrating DeepSeek’s success—OpenAI is now accusing them of illicitly “distilling” GPT’s knowledge and violating its terms of service. Isn’t that ironic? OpenAI pretty much scraped the web without permission, triggering multiple lawsuits. Now, they’re calling out DeepSeek for not respecting their terms of service.

Are we getting into a China vs. China race?

While OpenAI trades barbs with DeepSeek, another Chinese model—Kim EK1.5—stepped onto the stage, claiming even better performance than OpenAI’s “o1.” We’re now watching a surprising twist: China squaring off against itself for AI supremacy, leaving the U.S. worried it might get left behind. Meanwhile, here in Europe, we are busy fiddling with innovations like tethered bottle caps and clicking cookie banners. 

Why DeepSeek matters

DeepSeek isn’t just another run-of-the-mill chatbot clone:

• Open source: Everybody can look under the hood and can learn and modify the code.

• Chain-of-thought: It reveals each step in its reasoning process.

• Competitive performance: Early benchmarks suggest it can hold its own against bigger (and more expensive) models.

• Cheap: Cheaper than any other model with this kind of performance.

• Created without billions: We don’t need all that funding after all.

Maybe this will inspire Europe to enter the AI race, too. DeepSeek showed us that you don’t need all that capital to become a serious player in the market.🤷‍♂️

The future of open source

We already saw Meta making their AI model Llama open-source, which was a really good move. This built trust with developers, and many started to build on top of their open models. We also build our custom model for MindLumen, an AI therapist, on top of this open-source model LLama.

Open-source AI models like Llama and DeepSeek open the door for developers who don’t have bottomless budgets. And as more competition floods the market (especially from China), prices will likely keep dropping. Yes, there are concerns about censorship, security, and data use—but that’s the beauty of open source. You don’t have to use their AI clients directly; you can just download and run the code from your own computer. Locally, without even an internet connection. The easiest way to do that is to download Ollama and run ollama run deepseek-r1:1.5b in your terminal.

Is DeepSeek “stealing” from OpenAI’s tech? We don’t have definitive proof, but I wouldn't be surprised if they did. However, I wouldn’t be surprised if OpenAI didn’t get their data in a legit way either. Ultimately, this technology is here, and it’s not going anywhere. One thing is clear: open source is gaining ground, and if you’re building AI solutions, you may not need a $500 billion data center after all.