GPT-4.5 is touted as a model with 10% more mathematical and programming power, improving its emotional intelligence.
AI advancements continue: OpenAI has just unveiled GPT-4.5, an update that aims to improve the ChatGPT user experience through improvements in unsupervised learning and greater response accuracy, making it the company’s largest and most advanced model. The new model is only available to Pro users but should be available to ChatGPT Plus and Team users in the coming weeks.
GPT-4.5: A more precise and fluid model with fewer hallucinations
GPT-4.5 incorporates significant advances in information processing and response generation. One of the key improvements is the increased accuracy of its responses, which reduces so-called “hallucinations,” which occur when AI generates false or unreliable information.
To achieve this, OpenAI applied a renewed approach to unsupervised learning, which allows the model to identify patterns and understand context more deeply without relying on direct input. As a result, GPT-4.5 delivers more natural and conversational responses that facilitate more human-like interactions.
Furthermore, OpenAI has worked to strengthen its ability to understand emotional nuances and respond with greater emotional intelligence, adapting its tone to the context and mood detected in the user’s message. This aspect is crucial in applications such as customer service, creative writing, and digital coaching or consulting tasks.
Another relevant aspect is that GPT-4.5 is up to 10% more potent in logical reasoning, mathematics, and programming than -4. This makes it an ideal tool for solving complex technical problems while maintaining smooth interaction during daily consultations.
How to Use GPT-4.5
Upon its release, GPT-4.5 is available exclusively to Pro users, who pay a monthly subscription of $220. However, OpenAI has confirmed that it will soon be available to Plus or Team plan holders and later to businesses with ChatGPT Enterprise. ChatGPT Plus is priced at $22 per month and offers access to the GPT-4 model, while the free version works with GPT-3.5.
People with access to the model can select it directly from the model picker in the ChatGPT app, whether on the web, mobile, or desktop versions.
Artificial Intelligence with a Greater Focus on Security and User Experience
In addition to improved accuracy and fluency, OpenAI noted that GPT-4.5 incorporates security enhancements, reducing the likelihood of generating harmful or incorrect answers. The monitoring and tuning processes have also been refined, incorporating direct user feedback to improve response quality.
OpenAI explained that this version does not yet include multimodal features such as voice, video, or screen sharing. Still, the model supports working with files and images and integrates a digital canvas for collaborative writing and programming.
FAQS
OpenAI has introduced GPT-4.5, a significant advancement in its suite of extensive language models. Below are some frequently asked questions about GPT-4.5:
What is GPT-4.5?
GPT-4.5, codenamed Orion, is OpenAI’s latest primary language model, released on February 27, 2025. It represents a breakthrough in scaling pre- and post-training, primarily through unsupervised learning. This approach improves your ability to recognize patterns, make connections, and generate creative ideas without explicit reasoning.
How can I access GPT-4.5?
Since its release, GPT-4.5 has been available to ChatGPT Pro subscribers and developers via the OpenAI API. Access is expected to expand to other subscription tiers in the coming weeks.
What improvements does GPT-4.5 offer over previous models?
GPT-4.5 offers several improvements, including:
- Natural Conversation: The model is designed for more natural interactions, excelling at tasks that require creativity, empathy, and broad general knowledge.
- Emotional Intelligence: Demonstrates enhanced emotional intelligence, a better understanding of social cues, and nuanced communication skills, making interactions warmer and more intuitive.
- Accuracy: Users will experience fewer hallucinations and greater alignment with the original intent, resulting in more accurate and reliable responses.
What are the costs associated with using GPT-4.5?
Due to its large size and computational intensity, GPT-4.5 is more expensive to train and run. As of February 2025, via the OpenAI API, it costs $75 per million input tokens and $150 per million output tokens.
What are the limitations of GPT-4.5?
Despite its advances, GPT-4.5 is not a reasoned AI and may not outperform its competitors on some benchmarks. OpenAI recognizes that while GPT-4.5 offers more natural conversations, it is not a state-of-the-art AI model.
How does GPT-4.5 handle different languages?
GPT-4.5 was tested on the MMLU benchmark suite, which covers 15 languages, including Arabic, Bengali, Chinese, Korean, Spanish, French, German, Hindi, Indonesian, English, Italian, Japanese, Portuguese, Swahili, and Yoruba. The model outperformed its predecessor, GPT-4o, in all languages tested.
What are the plans for GPT-4.5?
OpenAI plans to expand access to GPT-4.5 to more users in the coming weeks and is working to integrate the model more seamlessly into various applications. The company aims to continue improving the model’s capabilities while addressing the tests associated with its computational requirements.