Google is rolling Gemini 4 Argon out on October 13, 2026. Google AI Pro and Ultra subscribers can access it in the Gemini app and AI Mode in Search; Google AI Ultra also has access in Antigravity. Developers can use the paid Gemini API. Those are different access routes, so an app subscription should not be treated as an API entitlement. Google’s launch announcement Gemini app update
Google describes Argon as a model for complex, sustained work, including coding and professional document tasks. This article uses a small fictional document case to teach the review step such work needs. It is not a test of Argon’s actual output or a comparison with another model.
Sources checked October 13, 2026. Product availability and claims below come from Google; the exercise and answer key were authored for AIViewer.
What is available today?
| Route | Google’s October 13 availability statement | What to check |
|---|---|---|
| Gemini app and AI Mode in Search | Rolling out to Google AI Pro and Ultra subscribers | Look for Argon in the model selector on your own account; a rollout may take time. |
| Google Antigravity | Google AI Ultra subscribers | The app announcement also describes Antigravity in the Gemini macOS app; account and platform requirements differ. |
| Gemini API | Paid API tier for developers | The documented model ID is gemini-4-argon. A Google AI Pro app plan does not itself establish paid API access. |
Google’s launch article gives the access routes, while its developer documentation specifies the API model ID and a unified context limit of 1,048,576 tokens. That developer limit is not a promise about what any particular Gemini app account can upload or generate. Google reports stronger performance on complex tasks, but its benchmarks and internal examples are company-reported evidence, not an independent AIViewer finding.
The same Gemini app update introduces a chat-to-tasks switch, more connected apps, and Antigravity in the macOS app. They are app changes announced alongside Argon; do not assume every connected action or coding feature is an inherent model capability. Google’s Gemini app update
A fictional three-document decision
All records in this section are fictional teaching inputs. Imagine a community group asking whether it can announce a room for a two-hour Friday workshop. The documents are deliberately short so every claim can be traced.
A — Event brief, current plan. The workshop needs a step-free room for 45 people on Friday, 3–5 p.m. The room budget is $200, including setup.
B — Venue email, sent after the first flyer draft. The North Room seats 50 and has a step-free entrance. It is available Friday, 3–5 p.m., at $75 per hour plus a $20 setup fee. The venue has not reserved it and asks for a reply to confirm.
C — Earlier flyer draft. “Friday, 4–6 p.m. The North Room is booked.” The draft was written before the current event brief and venue email.
Here is a deliberately flawed authored answer, not a Gemini output: “The North Room is booked for Friday, 4–6 p.m., seats 50 and costs $150, so send the flyer.” It mixes an outdated time and unconfirmed booking with an incomplete cost.
| Claim to review | Controlling evidence | Correct conclusion |
|---|---|---|
| Time | Current plan A and later venue email B agree | Friday, 3–5 p.m. The older flyer is stale. |
| Capacity and access | Venue email B | Seats 50 with step-free entrance, meeting the stated needs. |
| Cost | Venue email B | 2 × $75 + $20 setup = $170, which is $30 below the $200 budget. |
| Booking status | Venue email B explicitly says not reserved | Available is not booked. Someone must confirm with the venue before announcing it as final. |
A corrected answer would say: “The North Room appears to meet the current time, capacity, accessibility and budget requirements. The total quoted room cost is $170. The room is available but not yet reserved; confirm it with the venue and update the old flyer before sending anything.” The venue email controls the booking status because it is the venue’s later, direct statement; the flyer is a draft, not proof of a reservation.
Try the workflow on your own material
If Argon appears in your account, give it the documents with dates and labels intact. One bounded request is: “Answer whether this room can be announced as booked. Make a four-row ledger for time, capacity/access, total cost and reservation status. Quote or point to the controlling source for each row, show the cost calculation and list anything that still needs confirmation. Do not send or edit anything.”
Then compare the result with the downloadable source-check ledger. Check whether it keeps available separate from reserved, uses the newer source when documents conflict, includes the setup fee, and stops before an external action. If the answer is wrong, correct the specific row and inspect the full revised answer again. A confident response or a long reasoning trace does not replace that source check.
For real documents, use material you are permitted to share with the service and confirm any consequential action yourself. Google’s model documentation describes inputs and features; it does not certify that a generated answer will be accurate for your records.