Openai publishes GPT-4.5, in which you claim 10 times efficiency compared to GPT-4, but says that it is not a “border model”.

It is here: Openai has announced the publication of GPT-4.5A research preview of its latest and most powerful major language model (LLM) for chat applications doch, according to the company, especially in the system card of the model (a report in which its functions are performed online and is attached to the underside of this piece) that “GPT-4.5 is not a border model, but with OpenM-LLM enlargement of GPT-4 data model with an Openai countries.

It is also not a “argumentation model”, the new class of models that are from Openai, Deepseek, Anthropic and many other offers that produce the “chains of the thought” or the stream of consciousness, such as text blocks in which you reflect your own assumptions and conclusions to record mistakes before you convey answers/output. It is still a classic LLM. It is also amazing wide and away openai tax model (more on this below).

Nevertheless he is too The Openai co-founder and CEO Sam Altmans Post in the Social Network XGPT-4.5 reads: “The first model that feels talking to me with a thoughtful person. I had several moments when I leaned back on my chair and was amazed that they actually received good advice from a AI. “

However, he warned that the company joined the upper end of its range of graphics processing units (GPUS) and consequently had to restrict access:

“Bad news: It’s a huge, expensive model. We absolutely wanted to start plus and professional at the same time, but we grew a lot and are more out of the GPUS. We will add tens of thousands of GPUs next week and then roll it into the plus level. (Hundreds of thousands will come soon, and I’m pretty sure you will use anyone we can expand.)

We don’t want to operate like this, but it is difficult to perfectly predict growth fluctuations that lead to GPU bottlenecks.“

As of today, GPT-4.5 is available to the subscriber from Openai’s most expensive subscription levelChatgpt Pro ($ 200/month) and developers in all paid API levels with plans to expand access to the far more cost -effective plus and team stings (20 US dollars/$ 30 per month) in the next week.

GPT -4.5 can access the search and openai chatt -Canvas mode, and users can upload files and images, but still have no other multimodal functions such as language mode, video and screen sharing.

Openai is organizing a live stream event today at 12 p.m. PT/ 3 p.m. ET, at which Openai researchers discuss the development and skills of the model.

https://www.youtube.com/watch?v=cfryp0nitz8

Drive AI with unattended learning

GPT-4.5 represents a step forward in AI training, especially when learning unattended, which improves the ability of the model to recognize patterns, draw compounds and generate creative knowledge.

During the livestream, Openai researchers found how it was trained on data that was generated by smaller AI models, and that this improved its “world model”. They also indicated that it had been prepared via several data centers at the same time, which indicates a decentralized approach that resembles the nousn research of the competing laboratories.

This training regime has apparently helped to produce GPT-4.5 to produce more natural and intuitive interactions, to follow the user intent more precisely and to demonstrate greater emotional intelligence.

The model builds up earlier work on AI scaling on Openas, which increases the idea that increasing the data and calculating the service leads to better AI performance.

Compared to its predecessors, GPT-4.5 is expected to produce fewer hallucinations, which makes it more reliable for a wide range of topics.

What makes GPT-4.5 stand out?

According to Openaai, GPT-4.5 is developed to have warm, intuitive and of course flowing conversations. It has a stronger understanding of shade and context, which enables more human interactions and a greater ability to work effectively with users.

The extended knowledge base of the model and the improved ability to interpret subtle information enables it to highlight it in various applications, including:

• Writing aid: Refine the content, improve clarity and create creative ideas.

• Programming support: Debugging, proposal for code improvements and automation of workflows.

• Problem solution: Provision of detailed explanations and support in practical decisions.

GPT-4.5 also contains new realist techniques that improve its ability to understand human preferences and intentions and to further improve user experience.

So you use GPT-4.5

As of today, Chatgpt Pro-users can select GPT-4.5 in the model picker on web, mobile and desktop. Openaai will start with the introduction of plus and team users next week.

For developers, GPT-4.5 is made available via the Openai API, including the API of the chat degrees, the assistant API and the Batch API. It supports important functions such as functional views, structured outputs, streaming, system reports and image inputs, which means that it is a varied tool for various AI-controlled applications. It currently does not support multimodal functions such as language mode, video or screen release.

Pricing and implications for company decision -makers

Companies and team leaders benefit significantly from the skills introduced with GPT-4.5. With its improved reliability and natural conversation skills, GPT-4.5 can support a wide range of business functions:

• Improved customer loyalty: Companies can integrate GPT-4.5 into support systems for faster and more natural interactions.

• Improved generation of content: Marketing and communication teams can efficiently produce high-quality content in the fire fire.

• Optimized operations: AI-powered automation can contribute to the debugging, workflow optimization and strategic decision-making.

• Scalability and adaptation: The API enables tailor-made implementations so that companies can create AI-controlled solutions that meet their requirements.

At the same time the Pricing for GPT-4.5 by Openais API For developers of third-party providers who want to create applications on the model, with $ 75/180 per million input/output token, it appears shockingly high compared to USD $ 2.50 for GPT-4O.

And since other competing models have been released recently from the Claude 3.7 from Anthropic to Google’s Gemini 2 Pro from Google to Openais-owned argument “O” series (O1, O3-Mini-Hoch, O3)-The question of whether GPT-4.5 value is worth the relatively high costs, especially through API.

Early reactions from other AI researchers and power users vary strongly

The publication of GPT-4.5 has triggered mixed reactions of AI researchers and tech enthusiasts in the social network X, especially after the leak of the system card leak has resulted in standards before the official announcement.

Teknium (@teknium1)The pseudonymous co-founder of the rival AI model provider Nous Research expressed disappointment in the new model and pointed out minimal improvements in the MMLU results (multilingual understanding) and real coding benchmark compared to other leading LLMS.

“Since GPT-4 it has been used more than 2 years and 1000 times more capital … What happened?” he asked.

Others found that GPT-4.5 put in software engineering benchmark compared to the O3 mini model from Openai and raises questions about whether this publication is considerable progress.

However, some users defended the potential of the model beyond the RAW benchmark.

Software developer Haider (@slow_developer) The 10-fold improvement of the 10-fold computer efficiency of GPT-4.5 compared to GPT-4 and its stronger general skills compared to Openais Stem-focused O-Serie models from Openai was highlighted.

AI News Poster Andrew Curran (@Andrewurran_) Use a more qualitative perspective and predicted that GPT-4.5 new standards determine in writing and creative thinking and would describe it openais “opus”.

These discussions underline a broader debate in AI: Should progress only be measured on benchmarks or qualitative improvements in argument, creativity and human -like interactions have a higher value?

A research preview

Openai positions GPT-4.5 as a research preview to gain deeper insights into its strengths and borders. The company is still obliged to understand how users interact with the model and identify unexpected use cases.

“We share GPT-4.5 as a research preview to better understand the strengths and restrictions,” said Openai. “Scaling unattended learning continues to promote AI progress and improve accuracy, language reliability and reliability.”

Since Openaai continues to refine its models, GPT-4.5 serves as the basis for future AI progress, especially for argumentation and tool supply. While GPT-4.5 already demonstrates impressive skills, Openai actively evaluates its long-term role in his ecosystem.

With its broader knowledge base, improved emotional intelligence and more natural conversation skills, GPT-4.5 will offer significant improvements for users in various domains. Openai is excited to see how developers, companies and companies integrate the model into their workflows and applications.

While the AI ​​is developing, GPT-4.5 marks another milestone in Openas persecution according to more capable, more reliable and user-oriented language models that promises new opportunities for innovations in the corporate landscape.