Google Launches Gemini 4 Argon for Long-Running Agent Work
Google released Gemini 4 Argon on 13 October 2026, after first announcing it on 30 September. It is Google's flagship model for multi-step work that runs for a long time. Developers can call it through the paid Gemini API at an introductory price of $2 per million input tokens and $10 per million output tokens. That price doubles to $4 and $20 for work run from 1 December 2026. Google also added a tasks mode to the Gemini app, which keeps working when the user steps away and connects to more than 40 apps. Analysts quoted by InfoWorld advise budgeting at the post-discount rate, because per-token prices are a weak guide to what an agent costs per finished task.
In short
- Gemini 4 Argon became generally available on 13 October 2026 through the API, Google AI Ultra and Pro subscriptions, and the Gemini app.
- The introductory API price of $2 and $10 per million tokens doubles to $4 and $20 from 1 December 2026.
- The 1 million figure is a shared input and output window in the API, not a simple output cap.
- Analysts advise budgeting agent work at the post-discount rate and measuring cost per completed task.
- The new tasks mode connects to more than 40 apps but is limited to select countries for now.
What happened
Google put Gemini 4 Argon into general availability on 13 October 2026. According to Google's own launch post, developers and enterprises can use the model through the API in Google AI Studio and the Gemini platform. Google AI Ultra subscribers can reach it in Google Antigravity, its agentic coding environment, and Ultra and Pro subscribers can use it in the Gemini app and in AI Mode in Search. The model had been released first on 30 September 2026 to a small group of vetted cybersecurity partners through the Fairwind Program, according to TechCrunch.
Google describes Argon as built for hard, multi-step scientific, coding and agentic problems, where the model works for longer before answering. The main consumer change is a new tasks mode. According to SiliconANGLE, a single toggle in the Gemini app's main chat window switches the session into a mode where Gemini can keep working on a job after the user has left the computer. The same report says the app now connects to more than 40 third-party apps, including OpenTable, Shopify, Adobe and Instacart.
Why it matters for businesses that automate work
Most automation so far has been built around short requests: classify this email, extract these fields, summarise that document. Argon is aimed at the other end of the scale, where one instruction turns into dozens of steps across several systems. That is the kind of work that finance, operations and back-office teams already pay people to do by hand, such as reconciling records, chasing documents and preparing reports.
The pricing is the part that changes a budget. The introductory rates apply until 30 November 2026. According to a footnote in Google's post reported by InfoWorld, the rates rise to $4 and $20 afterwards. Google's API documentation and pricing page list cached input at a 95 percent discount, so the cached rate also doubles with the list price.
What changes in practice
The launch coverage contains one figure that is easy to misread. Some reports describe the output limit as rising from 64,000 to 1 million tokens. Google's developer documentation describes it differently. Input, thinking and output share one unified window of 1,048,576 tokens in the API. A single synchronous request still returns up to 65,536 output tokens, and longer output needs a background request that can run up to the full window minus the input.
For a business, the practical steps are these:
- Budget every agent task at $4 input and $20 output per million tokens, not the introductory rate.
- Measure cost per completed task in a pilot, because reasoning tokens are billed at the output rate.
- Review the permissions for every connected app before a task is allowed to place orders, book appointments or change records.
- Check long outputs in shared files, since a very large generated document is hard for one person to review in a chat window.
InfoWorld quotes analyst Sanchit Vir Gogia of Greyhound Research, who said that CIOs should budget at the higher rate from the first day, because no enterprise evaluation, security review or procurement cycle will finish within the discount window. He also argued that per-token prices say little about the cost of an agent, because the number of tokens a task uses is the real driver.
| Item | Introductory rate (to 30 November 2026) | Standard rate (from 1 December 2026) |
|---|---|---|
| Input, per million tokens | $2 | $4 |
| Output, per million tokens | $10 | $20 |
| Cached input, per million tokens | 95 percent off the input rate | 95 percent off the input rate |
Limits and what Argon does not do
Argon is not on the free Gemini plan, and Google's pricing excludes it from the Google AI Pro plan for API use. The Gemini app availability is limited to paying subscribers. Google's plans page, as reported by SiliconANGLE, says the tasks mode is limited to select countries for now, and that Argon is still rolling out to Pro subscribers. The benchmark results Google has published are the company's own figures, and independent testers have not confirmed every claim. A model that is strong at long tasks still needs a person to check the result before it reaches a customer or a ledger.
What to watch next
- The date when Argon's price changes, and whether Google extends the introductory window.
- Which countries receive the tasks mode, and when Pro subscribers receive Argon in full.
- Independent cost-per-task measurements for long jobs, which analysts have said are needed before budgets are set.
- Whether Google publishes the rest of the Gemini 4 family, since Argon is the only model so far.
How AiStaffo would automate this
A business that wants to put a long-running agent to work would typically start with one recurring back-office job, such as matching supplier invoices against purchase orders and chasing missing documents. AiStaffo would connect the accounting system, the shared inbox and the document store to the agent, so that the routine steps run automatically and every action is logged. The owner or finance lead still approves payments, reviews exceptions and decides which connected apps the agent may touch. Before go-live, AiStaffo would measure the cost of each completed task at the post-discount rate. Book a free automation audit.
Questions people ask
What is Gemini 4 Argon?
How much does Gemini 4 Argon cost through the API?
Is the 1 million token limit the same as output capacity?
What is the tasks mode in the Gemini app?
Should a business budget for Gemini 4 Argon at the introductory price?
Book a free automation audit
Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.
