An App Rises from Your Chat — Google Puts Gemini 4 Argon in the Gemini App and Pulls the Once-Separate Antigravity Right Next to the Conversation
On October 13, Google shipped Gemini 4 Argon to paid plans and integrated its Antigravity development platform inside the macOS Gemini app. We unpack the move to fold everything from building to testing into a single app, along with how to read its pricing and benchmarks.
On October 13, 2026, Google rolled out its latest model, "Gemini 4 Argon," to the Gemini app and to AI Mode in Search, and at the same time embedded its development platform "Antigravity" inside the Gemini app (macOS version). Argon, which had previously been limited to governments and a handful of partners (the Fairwind program), is now open to general users on paid plans as well.
A New Entry Point to Coding Inside the Conversation
What stands out is the Antigravity launch button added to the top right of the Gemini app (macOS). Press it and Antigravity's development environment opens right there, without leaving the app. The idea is that you can carry whatever you were discussing in chat straight into building an app or a web experience, all within the same app.
Antigravity is a development platform Google positions as "agent-first," and it has been offered as a standalone app since May 2026. With this update, its core coding agent interface now lives inside the Gemini app. The usage pattern looks like this: you spin up an agent in Antigravity, approve the plan it proposes, and from there code generation and testing proceed in the background, with a notification arriving once the work is ready for review.
The Numbers Argon Posted, and How to Read Them
Google is putting improved coding performance front and center for this model. On "DeepSWE v1.1," a benchmark that measures long-horizon software development, the comparison the company published is as follows.
| Model | DeepSWE v1.1 score (Google's published figures) |
|---|---|
| Gemini 4 Argon | 77.9% |
| Claude Opus 5.5 | 74.2% |
| GPT-6 Astra | 74.1% |
That said, it's worth keeping in mind that these are figures Google itself put forward. For internal use cases, the company says that agent-driven code submissions on Antigravity rose by more than 35%, rollbacks dropped by nearly 30%, and it handled refactoring on the scale of millions of lines in Chrome. All of this is best taken in the spirit of "such results have been reported." Third-party Artificial Analysis rates it as on par with GPT-6 Astra on its overall intelligence index, while noting that at introductory pricing, the cost per task is 60% lower.
Which Plan Unlocks What
Even within the same announcement, what you can actually use varies by plan. Using the Argon model and using Antigravity's development environment require different subscriptions.
| What | Plan required | Approximate price |
|---|---|---|
| Gemini 4 Argon (Gemini app / AI Mode in Search) | Google AI Pro or higher | About $20/month |
| Antigravity development environment (inside the Gemini app) | Google AI Ultra | From $100/month |
| Access via API | Usage-based pricing | $2 input / $10 output (per million tokens, introductory pricing) |
The API pricing is introductory and runs through November 30, 2026; after that it is set to double. In other words, the "cheap" part is only for the time being, and you'll need to estimate the cost of serious operation at the post-deadline price.
Beyond this, the Gemini app side also gained a "tasks mode" that lets work proceed even when you're not sitting in front of your PC (supporting over 40 integrations such as Instacart and Shopify), along with "Skills" invoked via @-mentions in place of slash commands, all added at the same time.
The Step of "Switching to Another Tool" Disappears
From a developer's point of view, what matters may be less the model's score and more the "unification of the entry point." Until now, it was standard to hammer out the design in chat, then switch to a separate IDE or CLI to implement it. This integration lowers the cost of that switch, folding everything from consultation to implementation and testing into the flow of a single app.
Moves in the same direction continue across the industry. Google has already extended Antigravity to VS Code and JetBrains, Cursor has been making its cloud agents resident, and OpenAI has been steering Codex toward a resident development platform. It's fair to read this as a shift in which the way of working — "hand an agent a goal, then review the finished result" — is becoming a shared default line across companies rather than a feature of any single tool. From a management perspective, which subscription you place as the entry point is starting to shape the day-to-day workflow of development teams directly.
What to Watch For Behind the Speed and the Savings
Alongside the convenience, there are points worth being careful about. First, the published coding performance is the vendor's own measurement, and it doesn't necessarily translate directly into real-world productivity. Second, a design where the agent drives everything from planning through implementation and testing in the background tends to pile up changes before a person has fully read the code, and some point out that the review burden becomes the new bottleneck. Third, serious agent development (Antigravity) presumes the $100+/month Ultra plan, and the introductory API pricing ends at year's end. It's safest to think of ease of trial and the cost of ongoing operation as two separate things.
References: 9to5Google: Google Antigravity is now built into Gemini app for macOS / SiliconANGLE: Gemini 4 Argon reaches Gemini app Pro and Ultra subscribers alongside a new tasks mode / Engadget: Google's Gemini 4 Argon is now available in the Gemini app and AI Mode in Search / 9to5Google: Gemini app rolling out Gemini 4 Argon to AI Pro and AI Ultra
