In plain English
- What happened
- Google DeepMind launched Gemini 4 Argon on October 13, 2026, making it available through the API in AI Studio and the Gemini platform, through Google Antigravity for AI Ultra subscribers, and in the Gemini app and AI Mode in Search for AI Ultra and Pro subscribers. The model launched at an introductory price of $2 per million input tokens and $10 per million output tokens for API use. Google says the model ranks first on the Vals Index, a benchmark covering finance, coding, legal and tax work.
- Why it matters to your business
- If a business already pays for Google AI Ultra or Pro, the newer model shows up in the same subscription at no extra cost, which is worth knowing before anyone assumes a separate upgrade is required. Anyone calling the model through the API, for a custom tool or integration, is billed per token at the new introductory rate, which Google has flagged as temporary.
- Do this
- Open the model selector inside the Gemini app or AI Studio and confirm which model is currently running on your account before relying on it for anything client-facing.
Gemini 4 Argon is Google DeepMind's newest frontier model, and it reached general availability across several products at once rather than a single staged rollout. The announcement describes it as built for deep, multi-step work in coding, finance, legal, cybersecurity and science.
Where it shows up and who can use it
| Where | Who can use it |
|---|---|
| API in AI Studio and Gemini platform | Developers and enterprises |
| Google Antigravity (coding tool) | Google AI Ultra subscribers |
| Gemini app and AI Mode in Search | Google AI Ultra and Pro subscribers |
For a small business, the practical version of this is simple: if the subscription is already Pro or Ultra tier, the new model is reachable without signing up for anything extra. Anyone on a free Gemini account is not included in this rollout based on what Google has published.
What it costs to call directly
Google set the API price at $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted 95%. Google describes this explicitly as an introductory price, which means it is not guaranteed to hold.
Google's own announcement calls the $2 input / $10 output per-million-token rate an introductory price, not a permanent one.
What Google says it is good at
Google reports that Argon ranks first on the Vals Index, a benchmark the company says measures performance across finance, coding, legal and tax tasks. The post also describes internal use at Google, including a large codebase refactor on the Chrome team and performance tuning on an open-source agent framework.
Those examples come from engineering teams with dedicated staff spending weeks on a single optimization project, which is a different scale than most small businesses operate at. The benchmark ranking and the internal case studies are both Google's own account of the model's performance, not independent testing.
Reality check
Most of the examples in Google's announcement involve research labs, large codebases and teams running the model autonomously overnight on narrow technical goals. None of that tells a solo business owner how the model performs on an everyday task like drafting a contract clause or reading a client's financial statement, and that is worth testing with real documents before trusting the output on anything that matters.
What to check before relying on it
- 1
Open the model selector
In the Gemini app, AI Mode in Search, or AI Studio, check which model is currently active on the account rather than assuming the newest one is default.
- 2
Confirm the subscription tier
Access to Argon in the Gemini app and AI Mode in Search is limited to AI Ultra and Pro subscribers according to Google's announcement, so a free account will not show it.
- 3
Test on a low-stakes task first
Run a document or question through the model that would not cause a problem if the answer were wrong, before using it for anything client-facing or financial.
- 4
Watch the API price if building anything custom
Google has labeled the per-token rate as introductory, so anyone wiring Argon into a tool through the API should expect the price to change.
Sources
