Google unveils Gemini 4 Argon at $2/$10 per 1M tokens, but access is limited to select users

Google unveiled Gemini 4 Argon on Wednesday with introductory pricing of $2 per million input tokens and $10 per million output tokens. That matches OpenAI's newly discounted GPT-6.1 Sol. Argon is not broadly available: Google is restricting initial access to select cybersecurity defenders and enterprise cloud customers while it participates in a voluntary U.S. government safety-testing process.

Key facts

• Model: Gemini 4 Argon, which Google positions as its new frontier model
• Introductory pricing: $2 per 1M input tokens, $10 per 1M output tokens
• Future pricing: Google says published rates will eventually double to $4 per 1M input tokens and $20 per 1M output tokens. No date for the change has been disclosed.
• Access: Select cybersecurity defenders and enterprise cloud customers only. Google plans to expand to developers, enterprise customers and consumers but has not announced a public release date.
• Context window, parameter count, training cutoff: Not disclosed in available reporting

Performance claims

Google is promising advances in coding, cybersecurity and complex tasks. According to industry benchmarks cited by CNBC, Argon ties OpenAI on a key cybersecurity test and posts leading results in software engineering. CNBC did not report specific scores, and the benchmarks have not been independently verified.

Independent verification is currently constrained by the restricted rollout. Bloomberg reported Wednesday that some Google employees questioned Argon's real-world coding performance despite strong benchmark results. Google disputes that characterization. It told CNBC that employees across the company have tested Gemini 4 versions for weeks, some with unlimited access.

Tulsee Doshi, Google's head of product for Gemini, highlighted Argon's ability to handle multi-step assignments that can run for extended periods. She said Google is still evaluating where to deploy it most effectively.

Agent strategy

The launch comes as investor attention shifts toward personal agents. Google's agent, Spark, launched in May at I/O and works across Gmail and Calendar. It can navigate websites through Chrome and complete tasks such as filling out online forms. It remains limited to paying subscribers. Spark cannot place outbound phone calls or complete purchases. It walks users through checkout and hands control back for final approval.

Meta's Muse app, by contrast, is free with usage caps. It had more than 5 million downloads as of Sept. 30, according to Sensor Tower, and reached the top of Apple's App Store, ahead of ChatGPT. OpenAI also rolled out its own personal agent, Dots, earlier this week.

Google is evaluating whether Argon could power more complex tasks within Spark. Doshi said the lighter, cheaper models Google released over the summer remain essential for everyday agent tasks. Frontier models like Argon would supply the extra reasoning needed for complex work.

Alphabet's stock is down about 6% over the past three months, while Meta is up 19%. JPMorgan analysts wrote Thursday that Google needs a significant advance in its personal agent offerings to generate consumer enthusiasm. Bank of America said Gemini 4 could strengthen Google's cloud business and existing products and provide a foundation for a future agent.

What this means

The $2/$10 price puts Argon level with OpenAI's discounted GPT-6.1 Sol, but the stated doubling to $4/$20 makes the introductory rate a temporary promotion. Teams evaluating Argon for long-running agent workloads should budget against the higher rate, since continuous multi-step agents consume output tokens heavily.

The restricted rollout means the benchmark claims cannot yet be checked by outside developers. The reported internal doubts about coding performance are disputed and unconfirmed, so real-world results will matter more than leaderboard positions once access widens.

The larger question is distribution. Google has the data surface in Gmail, Calendar, Chrome and Search, but Spark's paywall and lack of calling and purchasing give Meta's free Muse an adoption lead. Argon's strength on long, complex tasks fits agent use, but Google has not said it will power Spark, and the cost of running a frontier model continuously remains the constraint.