OPENAI releases the largest, most knowledgeable “GPT-4.5 model with reduced hallucinations and high cost of API
Join our daily and weekly newsletters for the latest updates and exclusive content of a leading AI coverage industry. Learn more
Here is: Openai announced the release of GPT-4.5Review of the study of the most powerful and powerful model with a large language (LLM) for chat applications. Unfortunately it is so The most expensive model of Openai (more about that below)S
This is also not a “reasoning model”, the new class of models offered by Openai, Deepseek, Anthropic and many others who produce “thought -out chains” or a stream of consciousness such as text blocks in which they reflect on their own assumptions and conclusions to try and serve the mistakes. It’s still more than a classic LLM.
Nevertheless, shares of OPENAI co -founder and the publication of Executive Director Sam Altman on the social network XGPT-4.5 is: “The first model I think to talk to a careful person with me. I have had several moments when I was sitting in my chair and was surprised that they were actually getting good advice from AI. “
However, he warned that the company encounters the upper end of the supply of graphic processors (GPU) and must limit access as a result:
“Bad News: It’s a giant, expensive model. We really wanted to release it on Plus and Pro at the same time, but we’re growing a lot and we’re out of graphics. Next week we will add tens of thousands of graphic processors and reach it on the smooth level then. (Hundreds of thousands are coming soon and I’m almost sure you will use anyone we can collect.)
So we do not want to work, but it is difficult to perfectly predict growth jumps that lead to a deficiency of a graphic processor.“
As of today, the GPT-4.5 is available to the subscribers of The most expensive subscription layer of OpenaiChatgpt Pro ($ 200 USD/month) and developers in all paid API Tiers, with plans to extend access to far less expensive plus and team levels ($ 20/$ 30 a month) next week.
GPT -4.5 is able to access the Cangpt Canvas mode of Openai and users can upload files and images to it, but there are still no other multimodal features such as voice mode, video and screen joke – after all.
Openai hosts a live event today at 12:00 PT/ 15 PM ET, where Openai researchers will discuss the development and capabilities of the model.
Improve AI with unsupported training
GPT-4.5 is a step forward in AI training, especially in unsupported training, which enhances the model’s ability to recognize models, draw connections and generate creative insights.
During Livestream, Openai researchers noted how to train data generated by smaller AI models and that this improved its “worldwide model”. They also said he was pre -trained in numerous data centers at the same time, suggesting a decentralized approach similar to that of competitive laboratory tests.
This training regime has obviously helped GPT-4.5 learn to create more natural and intuitive interactions, to follow the user’s intention, and to demonstrate greater emotional intelligence.
The model is based on the previous Openai work in the scales of AI, enhancing the idea that increasing data and computing power leads to better AI efficiency.
Compared to its predecessors, GPT-4.5 is expected to produce less hallucinations, making it more reliable in a wide range of topics.
What does the GPT-4.5 stand out?
According to Openai, GPT-4.5 is designed to create warm, intuitive and naturally flowing conversations. It has a stronger understanding of the nuance and context, which allows for more human interactions and a greater ability to cooperate effectively with consumers.
The extended base of the model knowledge and the improved ability to interpret fine signals allow it to be distinguished in various applications, including:
• Help to write: Content refining, improving clarity and generating creative ideas.
• Programming Support: Errors, offering code improvements and automation of working streams.
• Solving problems: Providing detailed explanations and support for practical decision making.
The GPT-4.5 also includes new alignment techniques that improve its ability to understand human preferences and intention, further improving consumer experience.
How to access GPT-4.5
Starting today, Chatgpt Pro users can choose GPT-4.5 in Picker Model on web, mobile and desktop. Next week, Openai will start deploying it on plus and team users.
For developers GPT-4.5 is provided via Openai’s API, including API to complete chat, API of Assistants and Batch API. It maintains key features such as functional calls, structured outputs, streaming, system messages and images inputs, making it a universal tool for various AI -controlled applications. At the moment, however, it does not maintain multimodal capabilities such as voice mode, video or screen sharing.
Pricing and consequences for the persons making decisions of the enterprise
Enterprises and team leaders take advantage of the opportunities introduced by GPT-4.5. With its increased reliability and natural conversational capabilities, GPT-4.5 can maintain a wide range of business features:
• Improved customer engagement: Business can integrate GPT-4.5 into support systems for faster, more natural interactions.
• Improved content generation: Marketing and communication teams can produce high quality brand content efficiently.
• Simplified operations: AI -powered automation can help to eliminate errors, workflow optimization and strategic decision making.
• Scalability and customization: API allows personalized conversion, which allows businesses to build AI-managed solutions suitable for their needs.
At the same time, Pricing for GPT-4.5 via Openai API For third-party developers who want to build model apps, it looks shockingly high at $ 75/$ 180 per million intense/output tortones compared to $ 2.50/$ 10 for GPT-4O.
And with other rival models recently launched from Anthropic’s Claude 3.7 to Gemini 2 Pro to Gues, to the Openai’s Own Series (O1, O3-Mini High, O3)-the question will be if the value of the GPT-4,5S costs, especially the high price, the value of the GPT.
Early reactions from colleagues from EI researchers and power users differ significantly
The issuance of the GPT-4.5 caused mixed reactions by AI researchers and technology enthusiasts on the social network X, especially after a version of the “system card” of the model (a technical document outlining its training and evaluations), which leaked earlier (included at the bottom of this article), revealing various results of the reference results.
Teknium (@Teknium1)The nickname co -founder of the competitive AI Nous Research models expressed disappointment with the new model, pointing to minimum improvements to MMLU (multilingual understanding) results and indicators of encoding in the real world compared to other leading LLMs.
“2+ years have passed and 1000 times more capital is located after the GPT-4 … What happened?” he asked.
Others have noted that the GPT-4.5 is lower than the OPENAI O3-Mini model compared to software engineering, raising questions about whether this version is significant progress.
However, some users have defended the potential of the model beyond the harsh performance.
Software developer Haider (@slow_developer) Frankly 10-fold improvement in the GPT-4.5 computational efficiency over the GPT-4 and its stronger general purpose than OPENA models focused on STEM.
AI News Poster Andrew Kuran (@Andrewcurran_) You have taken a better opinion, predicting that GPT-4.5 will set new standards in writing and creative thought, calling it “opus” to Openai.
These discussions emphasize a broader debate in AI: Do you have to measure progress in indicators or make quality improvements in reasoning, creativity and human interactions are more value?
Still in reviewing the research
Openai positions GPT-4.5 as a review of the study to get a more in-depth idea of ​​its strong and restrictions. The company remains committed to understanding how users interact with the model and identify unexpected use cases.
“We share the GPT-4.5 as a review of the study to better understand its strong and restrictions,” Openai said. “The scathing of incapable training continues to reach the progress of AI, improving accuracy, possession and reliability.”
As Openai continues to refine its models, GPT-4.5 serves as the basis for future AI progress, especially in agents for reasoning and using tools. While the GPT-4.5 already demonstrates impressive opportunities, Openai actively appreciates its long-term role in its ecosystem.
With its wider knowledge base, improved emotional intelligence and more natural conversational capabilities, GPT-4.5 is set to offer significant improvements to users in different domains. Openai is willing to see how developers, business and businesses integrate the model into their work processes and applications.
While AI continues to develop, GPT-4.5 marks another cornerstone in Openai’s pursuit of more capable, reliable and aligned language models, promising new innovation capabilities at Enterprise Landscape.