OpenAI GPT-4.5, Alexa+ and the move to wiser AI systems
With GPT-4.5 and Alexa+, generative AI systems are moving beyond just being LLM chatbots. Instead, they incorporate better human understanding and can perform tasks for humans.
AI systems released this week show that generative AI and AI agents are now part of life for consumers and businesses.
On Thursday, OpenAI released a research preview of GPT-4.5. OpenAI created GPT-4.5 by scaling unsupervised learning with compute and data, as well as architecture and optimization innovations.
The model was trained on Microsoft Azure, according to OpenAI. It has a better understanding of what humans mean and can interpret subtle cues or implicit expectations, the vendor said.
OpenAI's introduction of GPT-4.5 came a day after Amazon introduced Alexa+, a new generation of its AI assistant.
GPT 4.5 was also released on the same day that Chinese cloud provider Tencent released a new AI model that the vendor says can answer queries faster than DeepSeek-R1 and is comparable to DeepSeek-V3.
These different releases highlight a trend in which generative AI is moving beyond AI chatbot systems and standard LLMs to systems that can perform on behalf of users.
We're starting to see some very early stages of what is possible in terms of AI, more capably augmenting human workflows, or even potentially automating tasks that human beings used to do.
Arun ChandrasekaranAnalyst, Gartner
"We're starting to see some very early stages of what is possible in terms of AI more capably augmenting human workflows, or even potentially automating tasks that human beings used to do," said Gartner analyst Arun Chandrasekaran.
OpenAI GPT-4.5
For example, with GPT-4.5, OpenAI created a model with a broader knowledge and a deeper understanding of the world, which leads to fewer hallucinations, according to the vendor.
While not a reasoning model, GPT-4.5 feels natural and has an improved ability to follow user intent, according to OpenAI. This makes it better for writing, programming and solving practical problems.
In an attempt to differentiate its model from others, OpenAI is making it more human, said Andy Thurai, an analyst at Constellation Research.
OpenAI's claims that GPT-4.5's emotional intelligence, or "EQ," is better than others means that it attempts to respond with empathy just like a human would, Thurai said. "The interactions can feel more natural and human-like," he said.
However, there is also concern that it can fake emotional intelligence like humans sometimes do, Thurai continued.
The model is also expensive and restricted to Pro users for now. Pricing is $75 per 1 million cached inputs, $37.50 per 1 million tokens and $150 per 1 million output tokens.
OpenAI also released the system card for GPT-4.5. This is the second time this week that the vendor has released a system card explaining how it created a model; the first was for its deep learning agent Tuesday.
"This time, OpenAI voluntarily did this research and made the system card available, hoping enterprises will feel comfortable adopting their LLM over others in production environments," Thurai said.
With this, OpenAI appears to be taking on the challenges of governance, security, safety and privacy rather than avoiding them.
"By being transparent, they hope it will make more enterprises comfortable implementing their models in their production easily," Thurai added.
Amazon Alexa+
Another vendor that aimed to move generative AI systems beyond chatbot status is Amazon, with the revamped Alexa+.
According to Amazon, with generative AI, users can use Alexa+ to perform multiple tasks, including conversing with Alexa, receiving news summaries, managing calendars, creating and sending emails and texts, getting email summaries, ordering groceries and booking appointments.
"It's a big step forward," said Nikhil Lai, an analyst at Forrester Research. "Alexa's ability to interpret natural language is a big deal."
The new capabilities have long been expected in the digital assistant market, now dominated by Google, Apple and Amazon, Chandrasekaran said.
Amazon is integrating many capabilities with Amazon Alexa+, from reasoning and knowledge agents -- which can perform such tasks as making travel or restaurant reservations -- to computer usage with its ability to use an internet browser.
The capabilities will benefit marketers as they move away from keywords to natural words, Lai said. Instead of just targeting natural search keywords, marketers will use a more natural way of speaking to target their audience.
"This strengthens that possibility because Alexa is going to be retrieving information from various websites that you used to have to visit, that you're no longer going to visit," Lai continued.
However, Amazon must still figure out how to configure Alexa+ for enterprises. The vendor will also have to decide whether Alexa+ will be integrated with third-party apps or be part of a web portal, app or device, Chandrasekaran said.
"All of this will take them time," he said.
Moreover, Amazon competes with vendors that build AI agents and integrate them with native enterprise workflow apps such as Teams or Slack, he noted.
"They have a much more cohesive way of integrating with existing enterprise workflow and like user tools. Whereas, with this, Amazon has to fundamentally rethink how it's going to re-create that experience," Chandrasekaran said.
New Tencent model
Meanwhile, the Chinese AI industry continues to respond to DeepSeek's success, with Tencent being the latest AI vendor to introduce a new model in response.
Hunyuan Turbo S is not a reasoning model that needs to think for a while before answering, according to Tencent.
A model like this is valuable for real-time deployment of generative AI, Chandrasekaran said.
These are cases where AI systems are used for product recommendations and customer support in a real-time scenario in which the customer is on the line, and information is being gathered simultaneously, he said. However, he added that interest in this application is greater in China than in the U.S.
"I do not foresee other major providers at this point being interested in providing the Tencent model as a managed service," he said.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
