The most important thing in brief:
- DeepSeek:
- Powerful Chinese AI models
- Massively cheaper than established models from the USA
- New model (R2) to be released earlier than planned
- Critical Voices:
- Speculation: Export restrictions for high-performance GPUs circumvented?
- Copyright: Suspicion of using OpenAI models during training
- Privacy concerns
- prices: AI costs have fallen by a factor of 6—27 since March 2023
- Quality: The quality of AI models is increasing
- Education: Cheaper AI promotes equal opportunities and skills development
- Learn-With.AI: Integration of top models O1 and O3-mini by OpenAI
What is DeepSeek?
DeepSeek is a Chinese AI startup that develops large-language models that are comparable in performance to OpenAI (ChatGPT) — but with a fraction of the computing power and correspondingly lower costs. While the development of GPT-4 cost around 100 million USD, DeepSeek managed to implement a similar model for less than 6 million USD, according to its own statements. Due to US export restrictions, DeepSeek had no access to high-performance GPUs such as Nvidia's H100 and therefore had to develop particularly optimized training and inferencing methods (response generation). This increase in efficiency could help make AI technology significantly more affordable and accessible to the education sector.
Critical Voices
There are critical voices who claim that DeepSeek is secretly using Nvidia's high-performance H100 GPUs, despite US export restrictions that prohibit their delivery to China. According to Alexandr Wang, CEO of Scale AI, DeepSeek allegedly has around 50,000 of these GPUs, but this cannot be publicly communicated due to the restrictions.
There is also evidence that DeepSeek may have used OpenAI models to train its own AI. OpenAI and Microsoft are currently investigating this — but so far there is no definitive evidence.
Data protection is also viewed critically: DeepSeek stores user data on servers in China, which raises concerns about possible government access and the transfer of sensitive information. However, the model is open source and can be operated independently, so that it can be used without data transfer to DeepSeek.
Price development
The fact is that since the start of Learn-With.AI were able to see how prices for AI have fallen steadily while the quality was getting better and better. This trend is likely to increase even more now.
Since the release of GPT-4, prices per 1 million tokens generated (approx. 500,000 German words) have increased reduced by six times — from the original 60 USD to now just 10 USD. We expect that the costs of AI will continue to fall sharply. This is already evident today in models such as DeepSeek V3, which offers comparable performance to GPT-4o, but at just 1.10 USD, almost ten times cheaper is.
Specialized reasoning models developed for complex logical thinking and problem solving are also affected: While OpenAI o1 Still in January 2025 60 USD cost per 1 million tokens, offers DeepSeek R1, a comparable model, the same performance for just 2.19 USD — a reduction by a factor 27. OpenAI Has followed suit and its latest model O3-mini launched for 4.4 USD per 1 million tokens. In doing so, the company is responding to growing competitive pressure from China and is adapting its pricing strategy to remain competitive.
Good news for educational institutions
Access to powerful AI models has so far been restricted to economically better-off learners (ChatGPT Plus $21 and Pro $200 per month). Anyone with a paid Pro subscription not only has access to the best AI models, but can also use them more intensively, experiment more and thus build up valuable AI skills. This reinforces existing differences, as digitally savvy and financially privileged learners have a clear advantage — whether in individual learning or in exams with “open Internet” access.
Solution: Make AI accessible to everyone
To reduce this inequality, we have o1 and o3-mini from OpenAI directly in Learn-With.AI integrates. As a result, all learners and teachers receive from schools that are Learn-With.AI work, access to powerful AI models — regardless of financial opportunities. Schools and educational institutions can thus ensure that their learners benefit from cutting-edge AI, receive personalized support and build up AI skills in a targeted manner.
Falling costs — an opportunity for educational institutions
The falling costs of AI models are significantly reducing the financial hurdles for educational institutions. While only a small proportion of learners — mostly those with paid Pro subscriptions — have access to most powerful models , schools can provide this access for all learners.
This has two decisive advantages: On the one hand, Equal opportunities supported, as all learners — regardless of their social or financial background — can benefit from cutting-edge AI technology. On the other hand, wider access allows learners to use AI more intensively, experiment more and thus their Increasing AI competencies sustainably.
Reducing costs not only makes AI more affordable, but also builds decisive barriers From and opens new opportunities for the educational landscape.
Test now Learn-With.AI or contact us!
.png)
