google It may be back at the forefront of artificial intelligence, but as personal agents soar in popularity, investors and consumers are increasingly focused on what they can build with this technology.
The company unveiled its long-awaited Gemini 4 Argon model on Wednesday, promising significant advances in coding, cybersecurity, and complex tasks. According to industry benchmarks, argon OpenAI Participates in major cybersecurity tests and publishes outstanding results in software engineering.
Argon’s introductory price of $2 per million input tokens and $10 per million output tokens also aligns with OpenAI’s newly discounted GPT-6.1 Sol model. A token is about three-quarters of a word.
While Google becomes more aggressive in the model race, rivals meta has racked up millions of downloads with the Muse app, which was released last month and skyrocketed to the top. apple’s It surpassed the App Store and ChatGPT. Earlier this week, OpenAI unveiled Dots, a personal agent service, capitalizing on demand from businesses and consumers to provide tools that can perform tasks on your behalf, such as scrutinizing emails, booking travel, and organizing expenses.
On the other hand, safety is emphasized at the Frontier model level. human CEO Dario Amodei sparked a firestorm three weeks ago when he asked top AI labs to slow down their development amid growing concerns about potential risks.
Alphabet’s stock price has fallen about 6% in the past three months, but Meta has risen 19% in that time, following its steepest September rally since 2022.
analyst of JP Morgan Chase He wrote in a memo Thursday that Google needs to make significant advances in its personal agent service to capture consumer enthusiasm. bank of america We confirmed that Gemini 4 has the potential to enhance Google’s cloud business and existing products while providing the foundation for future personal agents.
In the consumer market, Google has a huge home court advantage.
Billions of people use Google products such as Gmail, Calendar, Chrome, and Search. Their messages and schedules are already stored within Google’s ecosystem, potentially giving the company a significant advantage in developing an Assistant that works across these services.
But Spark, Google’s personal agent, is still limited to paid subscribers, while Muse is free and has usage limits. As of September 30, Muse had more than 5 million downloads, according to Sensor Tower.
In a backgrounder with CNBC, members of Google’s agent team outlined Spark’s existing capabilities and the company’s approach to expanding access, acknowledging that Muse’s free use is helping consumers discover and experiment.
Samuel Boivin | Null Photo | Getty Images
Spark, which was announced at Google’s I/O developer conference in May, works across Gmail and Calendar, allowing you to navigate websites and complete tasks like filling out online forms through Chrome. It’s also available through Google’s mobile and desktop apps.
However, unlike Muse, Spark cannot make outbound calls or complete purchases on your behalf. Instead, guide the user through the purchase process before handing control back to the user for final approval.
Can argon help the spark?
Meta’s approach also comes with complexities.
Reuters reported in September that the company was experimenting with a human concierge system, with contractors handling some calls when Muse could not complete them autonomously. Meta has discontinued the experiment due to internal privacy concerns, but the automated phone call feature remains in beta.
This episode raised privacy concerns related to giving AI assistants access to so much sensitive personal information.
A Meta spokesperson said in an email that internal testing is “core to the product development process” and is conducted to obtain feedback on safety and privacy protections needed for broader release.
“We continue to work with our merchants to improve this potential calling feature and will only roll it out when it is ready and appropriate disclosures have been made,” the spokesperson said.
Google is currently evaluating whether the latest Frontier model can enhance Spark.
In an interview with CNBC, Tulsee Doshi, head of product at Gemini, highlighted Argon’s ability to tackle complex challenges that require multiple steps and run for long periods of time. He said Google is still evaluating where it can most effectively deploy argon.
Personal agents do not require the most powerful frontier models to complete their daily tasks. And before Argon, Google spent much of the summer releasing cheaper, lighter models designed to handle everyday requests at scale. Doshi said that while these models remain essential for everyday agents, more powerful frontier models can provide additional reasoning capabilities needed for complex missions.
I also have money. Running powerful AI models is expensive, especially for agents that work continuously and perform multiple tasks. Google’s Argon introductory pricing is competitive, but the company says its publishing fees will ultimately double to $4 per million input tokens and $20 per million output tokens.
One challenge for Google will be determining how to combine its models with agents that can handle increasingly sophisticated tasks without making the service cost-prohibitive to operate.
Argon is also not widely available yet.
Google is participating in the U.S. government’s voluntary safety testing process, initially restricting access to some cybersecurity advocates and enterprise cloud customers. The company plans to expand the offering to developers, enterprise customers, and consumers, but has not announced a general availability date.
This limits our ability to independently test Argon’s functionality. Bloomberg reported Wednesday that despite the positive benchmark results, some Google employees had doubts about the model’s real-world coding performance.
Google disputes this characterization, telling CNBC that employees across the company have been testing versions of Gemini 4 for several weeks, with some receiving unrestricted access.
But the urgency on the agent side is even greater as other companies race to bring AI to the masses.
“Google has struggled to translate its continued speed of product development into flashy releases that resonate with consumers, gain sustained mindshare, and meaningfully increase sentiment around AI leadership,” JPMorgan analysts wrote.
—CNBC’s Jonathan Vanian contributed to this report.
Featured: Google unveils new Frontier model called Gemini 4 Argon
