Google Opens Gemini 4 Argon to All Paying Users at $2 Per Million Tokens

Google has opened Gemini 4 Argon, the frontier AI model it announced weeks ago and initially restricted to cybersecurity defenders, to enterprises and paying subscribers worldwide. Access starts at $2 per million input tokens and $10 per million output tokens — introductory rates that double to $4 and $20 respectively after November 30.

The rollout, which Google DeepMind confirmed this week, puts Gemini 4 Argon in the hands of a far wider audience than the model's original release posture suggested. When the lab announced the model a few weeks ago, it said initial access would be limited to cybersecurity defenders. That restriction is now gone.

Where can developers actually use it?

Gemini 4 Argon is available globally across Google's main surfaces:

• Enterprises, through standard commercial channels
• Google AI Ultra and Pro subscribers, inside the Gemini app
• Android Studio, Google's mobile development environment
• Google AI Studio, the company's model experimentation platform
• Antigravity, Google's agentic developer platform for building autonomous AI systems (software agents that plan and execute multi-step tasks)

The pricing structure is the detail developers will notice first. The $2/$10 per million token rates are introductory. After November 30, the same usage costs $4 per million input tokens and $20 per million output tokens. For teams budgeting large-scale inference (the process of running a trained model on new inputs), that is a doubling of unit cost inside a single billing window.

What has Google actually done with it?

Google says it has already run Gemini 4 internally across software engineering, cybersecurity and research. In a blog post, the company detailed three concrete results:

• Gemini 4 has been used to optimize Python codebases
• The model identified and patched cybersecurity holes
• It made Google's open Gemma AI models three times quicker at tasks

That last figure is the sharpest number in the announcement. A 3x speedup on task completion, applied to a family of open models that third-party developers rely on, is a measurable internal validation — not just a marketing claim.

The model also played a role in developing physical AI, according to Google. It helped ground a humanoid robotics device in real-world physics and helped the system understand its surroundings. Google frames this as the model's core strength: reasoning about the physical world, mathematics and code rather than consumer-facing generative tasks.

Why does Google need a 'serious' STEM model right now?

The positioning is deliberate. Google is marketing Gemini 4 Argon as a hardcore companion for STEM workers — coding, mathematics, robotics, scientific research. That is a sharp turn from the era of Nano Banana, Google's AI image generator, which outperformed its competition and enraged creators whose work it was trained on and imitated.

The pivot tracks a broader industry shift. As criticism of AI and the companies building it mounts — from anger over data center construction to concerns about AI agents making unwanted incursions into users' systems and accounts — major technology companies have redirected their messaging toward AI's potential to solve fundamental problems, such as curing cancer. Google's emphasis on nuclear fusion research and robotics as Gemini 4 use cases fits that pattern exactly.

The competitive stakes are real. Gemini 4 Argon enters a frontier market against OpenAI's GPT-6 Astra and Meta's new Muse Wonder. Google held the lead among AI companies for most of 2025, but advancements from other labs have left it mostly behind this year. A successful Argon launch is Google's clearest opportunity to reclaim ground in a race it recently trailed.

What about the safety review?

Safety questions shadow the release. Concerns about AI safety have risen significantly over the past year, particularly around so-called "AI doomsday" scenarios — hypothetical outcomes in which advanced systems cause catastrophic, uncontrolled harm.

Google says it engaged with the White House's voluntary review process for new models and that Gemini 4 passed inspection under its safety framework. The company did not release details of that review. The process is voluntary and the public has no insight into its contents or rigor.

That opacity cuts both ways. It allows Google to claim regulatory engagement without exposing the evaluation to independent scrutiny. Critics of voluntary frameworks have argued this structure lets labs set their own bars. Google's position is that Gemini 4 met its own published safety standards before wider release.

"Only time will tell if Google's precautions are enough to prevent myriad AI-powered harms," as the company's own acknowledgment of residual risk frames the issue — a rare admission, in however hedged a form, that passing an internal review does not settle the question.

What should buyers do before November 30?

The practical calculus for enterprises is straightforward. Any team evaluating Gemini 4 Argon for coding, mathematical reasoning or agentic workflows has a window in which the model costs half its eventual rate. After the introductory period ends on November 30, input tokens double in price to $4 per million and output tokens to $20 per million.

For Google, the release completes a transition the company telegraphed from the start: from a narrowly gated cybersecurity tool to a globally available frontier model carrying Google's competitive hopes against GPT-6 Astra and Muse Wonder. Whether the internal results — the 3x Gemma speedup, the patched vulnerabilities, the grounded humanoid robot — translate into external demand at post-introductory prices will define Google's position in the frontier race through the end of the year.