Can Google’s Gemini 4 Argon Catch OpenAI and Anthropic?

TLDR

• Google released its new flagship AI model, Gemini 4 Argon, this week to compete with OpenAI and Anthropic.
• Benchmarks place Argon just behind Claude Opus 5.5 and Claude Sonnet 5.5 on a key industry leaderboard.
• The rollout is starting small, with access limited to cybersecurity partners and the U.S. government for safety testing.
• Argon’s pricing matches OpenAI’s discounted GPT-6.1 Sol model at $2 per million input tokens.
• Investors are more focused on personal AI agents, where Meta’s free Muse app is outpacing Google’s paywalled Spark.

---

Google introduced its newest AI model this week. It is called Gemini 4 Argon.

The company built the model to compete with top offerings from OpenAI and Anthropic. For most of 2026, those two companies have led the conversation around frontier AI systems.

Today we’re introducing Gemini 4 Argon.

It delivers frontier performance in complex workflows across real-world software engineering, knowledge work, and cybersecurity defense with an industry-leading 1M token output limit. pic.twitter.com/oDaCoOLu5E

— Google (@Google) September 30, 2026

Industry benchmarks show Argon performing well. The Artificial Analysis Intelligence Index ranks it just behind Claude Opus 5.5 and Claude Sonnet 5.5.

Analysts say the model shows strength in specific areas. These include legal reasoning, finance tasks, and long-running assignments for businesses.

Cybersecurity Is the First Focus

Google is rolling out Argon in phases. The first group to get access is trusted cybersecurity partners.

The company is also working with the U.S. government on safety evaluations before a wider release. Google says this approach lets it put a cyber-capable model into the hands of defenders quickly.

Analysts note this timing works in Google’s favor. Recent security problems tied to AI systems from other companies have created an opening for Google to position itself as a safety-focused option.

Argon also posts strong results on coding benchmarks. The model ties OpenAI on a major cybersecurity test and leads in software engineering scores, according to industry data.

Pricing for Argon starts at $2 per million input tokens and $10 per million output tokens. That matches OpenAI’s discounted GPT-6.1 Sol model. Google says those rates will eventually double.

The Real Competition Is in Personal Agents

While Argon strengthens Google’s position among frontier models, much of the market’s attention has shifted elsewhere. Personal AI agents are becoming the bigger focus for investors and everyday users.

Meta’s Muse app launched last month and quickly rose to the top of Apple’s App Store. It passed 5 million downloads by the end of September, according to Sensor Tower.

Google’s own agent, called Spark, launched in May. It can work across Gmail and Calendar, browse through Chrome, and fill out online forms.

Spark cannot make phone calls or complete purchases on its own. It hands control back to the user before any purchase is finalized. Muse offers more automation, though Meta paused a feature that used human contractors to handle some calls, citing privacy concerns.

Spark remains limited to paying subscribers. Muse is free with usage limits, which analysts say has helped it gain users faster.

Google is now studying whether Argon’s reasoning power could improve Spark. Product lead Tulsee Doshi told CNBC the model could help with tasks that require several steps over a longer period of time.

Analysts at JPMorgan Chase wrote that Google needs stronger personal agent products to win over everyday consumers. Bank of America said Argon could still help Google’s cloud business and existing products.

Over the past three months, Alphabet’s stock has fallen about 6%. Meta’s stock has risen 19% in that same period.

Argon is not yet available to the public. Google has not announced a release date for developers, businesses, or consumers.

Some Google employees have reportedly questioned how well Argon performs in real coding work compared to its benchmark scores. Google disputes this, saying employees have tested the model internally for weeks.

For now, Argon remains restricted to select partners while Google decides how to expand it into agent products.