Google is widening access to Gemini 4 Argon, and the rollout that began October 13 reaches Google AI Pro and Ultra subscribers, plus developers using the Gemini API. The model was first announced on September 30, and its launch pricing is half the eventual rate.
That gap is the part founders should care about. A tool that looks cheap this month can double your bill later, so the smart move is to budget for the higher number now. Our look at AI model pricing shows how quickly these costs can shift.
What Google Is Offering
Argon is positioned for complex, multi-step work such as coding and reasoning. It carries a context window of 1 million tokens, so it can take in long documents. Subscribers will see it as an option in the Gemini app and Search’s AI Mode, while Google’s agentic coding environment stays limited to Ultra subscribers.
Access has been staged, so test before you plan around it. One early reviewer reported an error when calling the model on October 1, which is a reminder that availability varies by account.
Coverage describes the model as aimed at demanding work such as coding and research. Those are broad claims, so a skeptical buyer should ask for evidence on the specific job they want to automate before moving real traffic.
The Price Tag, Before and After
The introductory rate is $2.00 per million input tokens and $10.00 per million output tokens. Cached input costs $0.10. Google’s footnote says the standard rate becomes $4.00 and $20.00 once the introductory period ends.
| Token type | Intro rate | Standard rate |
|---|---|---|
| Input | $2.00 | $4.00 |
| Output (including thinking) | $10.00 | $20.00 |
The end date is unclear. One analyst write-up points to November 30, while another coverage says Google has not named a date. Treat the promotion as temporary and plan for the standard rate.
In practice, that means building your forecast on the $4 and $20 figures and treating any savings as a bonus. If the promotion lasts longer than expected, you gain margin. If it ends early, you are already covered, and nobody has to explain a surprise invoice to your finance lead.
Why Cost Per Task Beats Cost Per Token
Per-token prices hide the real expense. Independent benchmarking from Artificial Analysis, as cited in a pricing review, puts Argon at about $1.99 per task at the intro rate and $3.98 at standard pricing. By comparison, GPT-6.1 Sol came in near $0.72 per task and GPT-6 Astra near $3.26.
The reason is verbosity. Benchmark runs show Argon writing around 62,000 tokens per task, against roughly 27,000 for GPT-6 Astra. A model that thinks longer bills longer, even when its per-token price looks competitive.
| Gemini 4 Argon, intro pricing | $1.99 |
| Gemini 4 Argon, standard pricing | $3.98 |
| GPT-6 Astra | $3.26 |
| GPT-6.1 Sol | $0.72 |
How to Budget Before You Commit
Start with a worked example. Say an agent run uses 200,000 input tokens and 60,000 output tokens. With cached input, that run costs about $0.62 at intro prices and $1.24 at standard prices. At 1,000 runs a month, your bill goes from roughly $620 to $1,240.
Caching matters here. Reusing the same instructions or documents across calls costs a fraction of fresh input, so build prompts with a stable opening section and a short variable section.
- Measure output tokens on your own tasks, not a vendor demo.
- Use caching for repeated context to cut input costs.
- Set monthly spending alerts before you ship.
- Keep a cheaper model ready for simple jobs.
Small teams exploring AI agents for small business should run the same math before choosing a vendor.
Quick Answers on Argon Access
Who can use Gemini 4 Argon now?
Paying Google AI Pro and Ultra members can choose it inside the Gemini app, and developers can reach it through the API and AI Studio.
Will the price really double?
Google’s footnote says the standard rate is $4 and $20 per million tokens after the intro period.
Is the long context window worth paying for?
Only if your tasks need it. Long inputs raise the bill on every call, so send the relevant pages instead of the whole archive.
Signals to Watch Before the Promo Ends
Watch for Google to name an end date, and for rivals to respond with price cuts or bundles. Competitive pressure has pushed prices down before, so a counter-move would not be a surprise.
Finally, track quality as well as cost. If a cheaper model handles your task nearly as well, the savings may outweigh a small accuracy gain.
Also test the new AI productivity tools from other vendors side by side. Takeaway: price the model on cost per finished task, and lock in your budget at the standard rate.






