Openai introduces the GPT-4.5 AI model in its research preview.

Openai released a survey preview of the GPT-4.5 Artificial Intelligence (AI) model on Thursday. The successor to the GPT-4O model arrives with natural conversation, reasoning and coding improvements. The AI ​​company called it “the largest and best model ever,” and emphasized that the company has expanded both pre-training and post-training with this model. Large-scale language models (LLM) have been developed using new technologies such as unsupervised learning along with traditional monitored fine-tuning (SFT) to develop reinforcement learning from human feedback (RLHF).

Openai has released the GPT-4.5 AI model

In a blog post, the San Francisco-based AI company announced the release of its GPT-4.5. This model is currently available as a research preview to help Openai better understand its advantages and disadvantages. Currently, only CHATGPT Pro subscribers have access to LLM, but the company says Plus and team users will get it by next week. Enterprise users and EDU users may then get it.

GPT 4.5 Benchmark Score
Photo credit: Openai

GPT-4.5 currently supports search, file, image upload, and canvas, but multimodal features such as audio mode, real-time video, and screen sharing in ChatGPT are not supported. For developers, the company previews new AI models with chat completion application programming interfaces (APIs), assistant APIs, and batch APIs for all paid tiers.

Interestingly, GPT-4.5 is not a frontier model that outperforms the older models in all metrics. The company’s internal benchmark tests show that the new model is better than the MMMLU (multi-language) O3-MINI and better than some coding-related benchmarks. “This isn’t a reasoning model and doesn’t crush the benchmark. It’s a different kind of intelligence,” Openai CEO Sam Altman said in a post on X (formerly known as Twitter).

Comparison of responses between GPT 4o and GPT 4.5
Photo credit: Openai

What the CEO meant was that the main focus of GPT 4.5 was to improve the understanding of AI’s conversational abilities and nuances. The company claims that the model “has a deeper understanding of human needs and intentions,” and now responds to conversation prompts in a more human-like way.

Its creative writing has also improved on a broader knowledge base. GPT-4.5 is said to come with a larger emotional quotient (EQ) that can closely follow the user’s intent and can provide better output through writing, programming and solving real problems. It is also claimed to have fewer hallucinations than previous models.

Furthermore, GPT-4.5 also improves inference capabilities. In particular, it is not a reasoning model and does not depend on test time calculations, which increases processing time and improves response accuracy and depth. Instead, Openai said that LLM shows enhanced inference as a core feature, providing better responses in real time.

These features are said to be added by scaling existing methods and implementing new methods at both pre-training and post-training stages. Openai used unsupervised learning to increase the accuracy and intuition of the world model. Additionally, SFT and RLHF technologies have also been extended with this model.