In a world where AI models are pushing new boundaries almost weekly, OpenAI unveiled its latest flagship product on February 27, 2025. With ChatGPT 4.5, the company is pursuing a remarkable shift in direction: instead of technical superlatives, human interaction is taking center stage. But what does this shift in perspective mean for the future of AI?
From robotic helper to “AI buddy”
"It's the first model that feels like talking to a thoughtful person," OpenAI CEO Sam Altman explained in a post on X shortly after the release. This statement may initially sound like typical Silicon Valley marketing, but early user reports suggest there's more to Altman's words than mere rhetoric.
The model, known internally as "Orion," was initially released as a "Research Preview" for the most affluent users—subscribers to the $1,400-per-month ChatGPT Pro plan. Plus users will have to wait until next week before receiving access.
Dan Shipper, CEO of AI and the business newsletter Every, aptly describes the update: "Not a quantum leap in intelligence, but rather an upgrade in personality, communication, and creativity. It's as if OpenAI is evolving its base model from 'boring assistant' to 'AI buddy.'"
A step towards emotional intelligence
The most impressive innovation of GPT-4.5 is undoubtedly its improved emotional intelligence. While previous AI models routinely interpreted queries literally, GPT-4.5 can grasp the emotional nuances of a conversation and respond accordingly.
In one telling test example, the model was asked to compose an angry message to a friend who had canceled plans. Instead of blindly following the instruction, GPT-4.5 recognized the underlying frustration and instead suggested a more constructive response—a subtle but significant difference from the mechanical compliance of previous generations.
This ability to read between the lines and consider emotional context makes GPT-4.5 particularly valuable for applications that require empathy: from customer support to educational aids to creative collaborations.
Conciseness instead of information overload
Another qualitative leap is evident in the model's communication skills. Unlike its predecessors, which often tended toward lengthy explanations, GPT-4.5 provides precise and understandable answers.
In a comparative test on the simple question "Why is the sea salty?" GPT-4 presented a comprehensive yet overwhelming discussion, while GPT-4.5 offered a clear, concise, yet complete explanation. This ability to be concise without losing information represents a significant advance in usability.
The technical basics
Behind the user-focused improvements lie impressive technical advances. GPT-4.5 offers:
- Ten times higher computational efficiency compared to GPT-4
- Significantly reduced “hallucinations” (false or fabricated information)
- Improved context processing and knowledge base
- More natural conversation flows with less robotic response patterns
These improvements are especially notable considering that OpenAI calls GPT-4.5 its “last non-chain-of-thought model” before fully transitioning to structured reasoning.
The other side of the coin
Despite all the progress, GPT-4.5 comes with significant limitations. Its greatest weakness lies precisely in its strength: its focus on natural communication and emotional intelligence comes at the expense of structured logical skills.
In benchmarks for scientific and mathematical reasoning, GPT-4.5 lags significantly behind OpenAI's specialized O-series models:
| Benchmark | GPT-4.5 | GPT-4o | OpenAI o3-mini |
|---|---|---|---|
| GPQA (Science) | 71,4% | 53,6% | 79,7% |
| AIME '24 (Mathematics) | 36,7% | 9,3% | 87,3% |
| MMLU (multilingual) | 85,1% | 81,5% | 81,1% |
“It’s not going to break any benchmarks,” Altman himself admitted, calling it a “huge, expensive model”—a surprisingly honest assessment in an industry known for its hyperbole.
The question of price
Perhaps the biggest controversy surrounding GPT-4.5 concerns its price. AI researcher Aran Komatsuzaki noted that API access to GPT-4.5 is about 15 to 20 times more expensive than GPT-4o—at $0.75 per million input tokens and $0.150 per million output tokens, it represents a significant investment.
Ashutosh Shrivastava, founder of the AI Compass newsletter, summed up the frustration of many: "OpenAI GPT-4.5's pricing is insane. What are they thinking?"
These high costs create a problematic barrier to access, especially for researchers and smaller institutions that could benefit from the model's advanced capabilities but lack the necessary resources.
An intermediate step on the way to GPT-5?
OpenAI positions GPT-4.5 as a bridge between its current capabilities and the highly anticipated GPT-5. The full stable release of GPT-4.5 is expected in Q4 2025, suggesting that the current preview version still has room for improvement.
In the long term, OpenAI will evaluate whether to retain GPT-4.5, depending on user acceptance and feedback. This cautious approach underscores that the company itself views GPT-4.5 as an experimental foray toward more human-like AI.
Conclusion: Evolution instead of revolution
ChatGPT 4.5 marks a significant, albeit not revolutionary, step in AI development. Its strengths lie less in quantifiable leaps in performance than in the qualitative improvement of human-machine interaction.
At a time when AI systems are increasingly integrated into our everyday lives, this focus on personality and emotional intelligence could prove far more valuable than pure technical benchmark successes. The true innovation of GPT-4.5 is that it feels less like a machine and more like a conversational partner—a subtle but significant step toward an AI that is not only intelligent but also approachable.
Whether this new direction justifies the high price and whether OpenAI has found the right balance between emotional and logical intelligence will become clear in the coming months. One thing is certain, however: with GPT-4.5, OpenAI has significantly raised the bar for natural AI interactions.
We'll consider in detail whether to make ChatGPT or the API version available for SaaS or other areas of the website, but as always, practicality comes before hype. We'll try to cover this with our usual high quality in our review and future articles.