Last week, OpenAI announced GPT-4.5, describing it as its “largest and most knowledgeable model yet.” Initially, it was exclusively available to ChatGPT Pro subscribers for $200 per month. However, today, it has become accessible at a lower cost.
On Wednesday, OpenAI announced via X (formerly Twitter) that GPT-4.5 is now available to ChatGPT Plus users for $20 per month. While the full rollout was expected to take one to three hours, it was completed within an hour. OpenAI has not yet provided details on specific usage limits but noted that rate limits will be adjusted based on user demand.
>>>C926647400P for BLU C926647400P
What Is GPT-4.5?
GPT-4.5 aims to offer better alignment with user intent, fewer hallucinations, enhanced emotional intelligence, and more natural interactions. These improvements stem from unsupervised learning and enhanced reasoning capabilities.
Unlike OpenAI’s “o1” model, GPT-4.5 does not utilize chain-of-thought (CoT) reasoning, but it delivers efficient reasoning and improved social cue awareness.
During demonstrations, GPT-4.5 responded more intuitively than the “o1” model. For instance, when asked to generate hateful content, GPT-4.5 provided nuanced alternatives rather than outright negative responses. Additionally, for technical inquiries, GPT-4.5 offered more natural and fluid responses compared to the structured answers given by “o1.” This makes GPT-4.5 well-suited for everyday tasks such as writing and problem-solving.
GPT-4.5 incorporates traditional supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), along with new supervision methods. In a comparative demonstration of different GPT models answering the question, “Why is seawater salty?” GPT-4.5 stood out with engaging and conversational responses.
GPT-4.5 integrates advanced ChatGPT features such as Search, Canvas, and file/image uploads. However, it does not yet support multimodal features like Voice Mode or video. OpenAI plans future improvements to make transitioning between models more seamless.
>>>SLC-51 for Samsung Galaxy A05S
Benchmark Performance
GPT-4.5 outperformed its predecessor, GPT-4o, in key benchmarks such as Competition Math (AIME 2024), PhD-level Science (GPQA Diamond), and coding tests (SWE-Bench).
Notably, GPT-4.5 closely matched or even surpassed OpenAI’s reasoning-focused model, “o3-mini,” in specific areas, particularly in coding (SWE-Lancer Diamond) and multilingual (MMMLU) benchmarks.
Additionally, according to SimpleQA evaluations, GPT-4.5 demonstrated higher accuracy and fewer hallucinations compared to previous models, including GPT-4o, o1, and o3-mini.
In human comparative testing, GPT-4.5 was preferred over GPT-4o for everyday tasks, professional use, and creative work.