Google shows it's not out of the AI race just yet

By Emily Bary and Christine Ji

The company's Gemini 4 Argon announcement shows Google is back with the pack when it comes to cutting-edge AI models

Google offered a preview of its Argon model on Wednesday.

It wasn't too long ago that Google parent Alphabet was viewed to have arguably the top artificial-intelligence model in the industry. Lately, though, investors have come to worry about delays in the rollout of the company's updated model, all while rivals like OpenAI and Anthropic have come out with cutting-edge offerings.

So Google's (GOOG) (GOOGL) announcement Wednesday of its new Gemini 4 Argon model comes as welcome news. In a blog post, the company said the model "delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense."

The rollout of Argon won't be immediately broad. Recent concerns about AI agents going rogue and carrying out undesirable actions have lead to discussions among AI leaders about slowing down the pace of model development.

"Safely releasing frontier capabilities at this level requires a phased approach," Google said in the post.

The company intends to "gradually expand access" after voluntarily participating in a government program that has cybersecurity experts vet models before their official release.

Google also shared that it was deploying misalignment mitigations to monitor Argon's chain of thought and shut down actions when necessary. It is also taking increased precautions with its training sandboxes to prevent agents from breaking out of secure testing systems.

"We strongly encourage the rest of the industry to preserve reasoning transparency in these pivotal moments of increased capabilities while navigating alignment risks," Google said.

The company touted Argon's performance benefits relative to its previous model, and offered its own data suggesting that Argon is better than OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 when it comes to various benchmarks related to knowledge work and AI coding assistance.

That said, judging AI models gets complicated. "While much remains to be seen, its clearest differentiator is its industry-leading output allowance of 1 million tokens," Evercore ISI analyst Mark Mahaney wrote in a note to clients. He flagged that Google is starting off by pricing Argon at $10 per million output tokens and $2 per million input tokens.

Tokens are a unit used to measure what AI models are able to receive and process. Input tokens cover what a user submits in a query, while output tokens represent what the model generates in response. Google's pricing scheme will allow customers to run more complex workflows and receive more output per dollar compared to other models.

The pricing structure is highly competitive, according to Mahaney. "This introductory pricing undercuts GPT-6 Astra and Anthropic's premium models, while matching Grok 4.7 on short-context input pricing," he wrote, referring to SpaceX's (SPCX) Grok model. However, Google said its pricing levels will double from the introductory rate over time to be on par with what Anthropic's Claude Opus 5.5 uses.

Since Argon is undergoing a limited rollout at the outset, it won't contribute to meaningful financial gains for Google right away. The significance of the release is more symbolic, Mahaney suggested.

"We think the announcement reinforces the breadth of Google's AI platform and its ability to compete at the frontier," he wrote.

Alphabet's stock increased 1% in Wednesday trading and was up another 1.5% in the extended session.

-Emily Bary -Christine Ji

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

09-30-26 1851ET