Chinese Pressure on the Global Market: DeepSeek and Alibaba Change the Rules
In January 2025, the world of technological engineering paused when Chinese startup DeepSeek introduced its R1 model. This move was not just another model release; it was a signal that efficiency and openness can compete with the enormous investments of American corporations. Shortly thereafter, Alibaba followed, introducing the model QwQ-32B, which, with its logical reasoning capabilities, is on par with top models like DeepSeek-R1.
For comparison: While models like GPT-4o from OpenAI or Claude 3.5 Sonnet from Anthropic excel in their versatility and immense contextual ability, Chinese models focus on extreme efficiency in logical reasoning and programming. This means that at much lower training and operating costs, they can achieve results that were previously reserved only for the largest players with multi-billion dollar budgets.
Comparative View: Benchmarks and Performance
If we look at benchmarks in mathematics and programming, models like DeepSeek-R1 show capabilities that are in direct competition with the best versions of GPT. While OpenAI maintains a lead in creative writing and complex ecosystem integration (e.g., Microsoft Copilot), Chinese models offer developers something fundamental: openness. This allows companies to run these models on their own hardware, which is crucial for security and data control.
Strategic Shift: Sam Altman and the End of the Era of Closed Models?
This pressure from the East also impacts the leaders of the American AI sector themselves. According to reports from the Wall Street Journal, OpenAI CEO Sam Altman admitted that the company is on the "wrong side of history" regarding open source. The traditional "closed-source" model, where the user is dependent on the provider's API, is starting to lose ground in confrontation with a community that can fine-tune models to their own needs.
Microsoft is now considering whether to replace some parts of its Copilot with its own models to reduce dependence on OpenAI. We are thus seeing a process of market fragmentation, where new alliances are forming and old ties are breaking down. For the Czech user, this means a wider selection of tools, but also a greater need to understand what model their software is actually working with.
On-Device AI and Vertical AI: The Path to Efficiency
Part of this shift is also the trend towards so-called On-Device AI and Vertical AI. As market analyses indicate, the future is not just in giant clouds, but in intelligence that runs directly on your device (laptop, phone, or industrial sensor) without the need for an internet connection.
- On-Device AI: The model runs locally thanks to specialized chips. This ensures maximum privacy – your data never leaves the device.
- Vertical AI: Models developed for a specific industry (medicine, law, manufacturing automation). They are less universal but much more accurate and reliable in their given field.
This trend is very relevant for the European market, where regulations like the EU AI Act are strict on privacy protection. The ability to use local models (e.g., using tools like Ollama) allows companies in the Czech Republic to meet the strictest GDPR standards without having to send sensitive information to American or Chinese data centers.
What Does This Mean for the Czech Market and Companies?
For the Czech IT sphere and small and medium-sized enterprises (SMEs), this situation brings several fundamental opportunities and challenges:
- Cost Reduction: Using open-source models (DeepSeek, Llama, Qwen) can drastically reduce the cost of operating AI services compared to a ChatGPT Plus subscription for $20/month or expensive OpenAI API calls.
- Data Sovereignty: Czech companies can implement AI within their own IT infrastructure thanks to open models, which is crucial for public administration, banking, and healthcare.
- Availability in Czech: Although these models are primarily trained on English, their ability to understand languages is significantly increasing thanks to methods like "instruction tuning." For the Czech market, it is important to monitor how well these models handle our specific grammatical forms in local deployment.
Pricing Policy: While premium services like ChatGPT Plus or Claude Pro cost around $20 (approx. 470 CZK) per month, the use of open-source models is essentially free in terms of licensing. The only costs are hardware or cloud computing costs, which for smaller models can be just a few crowns per million tokens when using an API.
Is it safe to use Chinese models like DeepSeek for sensitive data in the Czech Republic?
If you use the model via a cloud API directly from a Chinese provider, data travels to China. However, because these models are open-source, you can download and run them on your own server in the Czech Republic or within the EU. In this case, your data is fully under your control and meets GDPR requirements.
How well do these new models understand Czech?
Their level is constantly improving. Although they are not natively "Czech" models, thanks to the enormous amount of available texts on the internet, they handle Czech grammar and context very well, especially in tasks like summarization or programming. However, for extremely specific Czech legal or medical terms, it is still better to use models with local fine-tuning.
Can a small Czech company compete with large players thanks to these models?
Yes, precisely because of them. Open-source models democratize access to cutting-edge technology. A small company does not have to spend billions on developing its own model; it just needs to invest in implementing and specializing (Vertical AI) an existing open-source model for the specific needs of its customer.
