OpenAI has unveiled its most advanced artificial intelligence model yet – GPT-4.5, codenamed “Orion”. A curious incident followed its release: initially, the technical documentation included a statement saying the model was “not cutting-edge.” However, after publication, OpenAI decided to remove this line, likely acknowledging that the model is indeed a breakthrough.
Who is it for?
GPT-4.5 has been built using more extensive datasets and powerful computational resources compared to previous versions. OpenAI emphasizes that this model is not meant to replace GPT-4, but rather to serve professional users. It is currently available for ChatGPT Pro subscribers and OpenAI API users, while ChatGPT Plus subscribers are expected to gain access in the coming days.
GPT-4.5 and its “Emotions”
According to OpenAI, GPT-4.5 understands human conversations better and responds in a more natural and warm tone than its predecessors. It is also significantly better at grasping user intentions, making interactions feel smoother and more intuitive.
In one experiment, GPT-4.5 was the only model that successfully created an SVG image of a unicorn using mathematical formulas and code.
Accuracy and coding capabilities
GPT-4.5 excels in answering factual questions. In SimpleQA tests, it outperformed GPT-4o and other models, including o1 and o3-mini. This means it is less likely to generate false or misleading information, a major improvement in AI reliability.
Its coding skills are also impressive. In SWE-Bench Verified and SWE-Lancer tests, GPT-4.5 performed on par with GPT-4o, though it lagged behind top-tier coding models like Claude 3.7 Sonnet.
Limitations and pricing
Despite its impressive capabilities, GPT-4.5 has some limitations. For example, it does not support real-time two-way voice interactions, a feature available in models like GPT-4o.
Additionally, the model requires high computational power, which significantly affects pricing. According to OpenAI:
- $75 per million input tokens
- $150 per million output tokens
This is far more expensive than GPT-4o, which costs only $2.50 per million input tokens.
Challenges in scaling up
Even though GPT-4.5 has been trained on a vast dataset, experts believe that this model may have reached its limits in terms of scalability. The future of AI will likely focus not just on expanding data, but also on developing more advanced reasoning and decision-making capabilities.
OpenAI has already outlined its future direction – GPT-5 is expected to be released by the end of this year, integrating AI models with more sophisticated reasoning abilities.
Conclusion
GPT-4.5 brings significant improvements in accuracy, emotional intelligence, and creativity. However, it is likely just a stepping stone toward more powerful AI systems.
Its precision, creative potential, and deeper understanding make it an excellent tool for advanced users. However, its high cost and certain limitations may prevent it from becoming the perfect model for everyone.
