Summarize & listen to article
Artificial intelligence is undergoing a significant transformation. Early 2025 shows new players entering the market with powerful alternatives to existing (generative) AI services. What does this mean for users and businesses?
New AI models from china
China is in the mix without having access to the heavy NVIDIA chips. DeepSeek has achieved a breakthrough with their new reasoning model. This freely available model demonstrates that powerful AI technology is no longer exclusive to major Western tech companies. Further more, it was trained for significant lower costs and also runs at lower costs. While I don’t believe everything that is said, it is made much cheaper then the closed modals of Anthropic and OpenAI.
Most likely this will have an impact on how the big tech companies will develop and train their modals. Why do we have to pay so much money if it can be much cheaper?
Key features of DeepSeek:
- Completely free to use
- Developed at lower costs
- More efficient in operation than many existing models
- Performance comparable to premium services
Alibaba has also joined this development with Qwen2.5-Max. This model can compete with well-known players like ChatGPT and Claude Sonnet 3.5, demonstrating that Chinese AI technology has become a serious competitor.
How is OpenAI responding?
OpenAI is developing on two fronts. First, there’s the expansion of o1 with canvas support, enabling users to write, code, and analyze data more effectively. Additionally, OpenAI has quietly introduced a new development: the think-switch function. Via Perplexity search I found out that now, the default model seems to be o3-mini and is switched to o1 when the “Think” button is clicked.

What’s new at OpenAI?
- Canvas integration in o1 for enhanced basic functions
- Think-switch for access to advanced reasoning
- o3-mini as the standard underlying model
- Focus on practical applicability
The rise of AI-agents
Well, not really yet. AI-agents the way I like them to see, are autonomous, smart pieces of software that can reason, decide, change and improve themselves when necessary. They are context-aware and make decisions on their own without a human in the loop. And that’s not the case for today’s agents. They are more automation programs, but that’s fine.
Perplexity’s new Android assistant is one of the new kids on the block. This tool goes beyond simple voice commands and can actually perform tasks across different apps.
What can the Perplexity Assistant do?
- Book Uber rides
- Play music through Spotify or YouTube
- Make restaurant reservations
- Maintain context between different actions
- Respond to camera input
- Answer questions about screen content
Meanwhile, OpenAI is testing their Operator agent. Currently available in preview only to American PRO users for $200 per month. I saw a couple of videos and it is kind of disappointing. Within a chat you will see a browser window and Operator is moving the cursor and doing stuff in the browser.
Important Considerations for Users
The new developments offer more choice but require conscious decisions. When using free services, it’s particularly important to consider data usage and privacy.
What to watch out for:
- Free models store user data
- Chinese models use servers in China
- Balance between costs and data privacy
- Functionality versus security
What Does This Mean for the Future?
These developments show that AI technology is becoming increasingly accessible. The combination of free models and practical AI assistants opens up new possibilities for both individuals and organizations.
At the same time, choices are becoming more complex. It’s no longer just about what features an AI tool has, but also about questions surrounding data usage, privacy, and responsible deployment of the technology.