Gemini 4 Argon: what changes for app users and developers

Google has released Gemini 4 Argon, its new artificial intelligence model, in the Gemini app and in its API for developers. In the app, the model has been available to Google AI Pro and Ultra subscribers since the announcement, according to the company. Google introduced the app features and the model's safeguards on Oct. 13 and updated the API documentation on Oct. 14. All three items appear in the Oct. 14 edition of AI Updates, a daily AI newsletter.

In the app: delegated tasks and links to other apps

In a post on its blog on Oct. 13, Google said Argon gives more direct answers, with fewer caveats and less repetition, and reports progress on long tasks more often. Those are the company's own words.

The app also gets a Tasks mode, a toggle in the chat window that lets users hand work off to Gemini, even while they are away from their computer. Google says those tasks can run in its own apps and in third-party ones. The app now connects to more than 40 other apps; the company named Instacart, OpenTable, Adobe and Shopify as examples. Antigravity, Google's agent-based coding interface, is now available in the macOS app.

For people using Gemini in Brazil, Google's announcement does not say which countries and languages get the new features, whether the free plan has access or what the usage limits are. Nor does it confirm a launch in Brazil. The app itself is the place to check what is turned on for each account.

In the API: two price tiers by date and breaking changes

Google's developer documentation, updated Oct. 14, lists two sets of prices for the gemini-4-argon model, depending on the date. Through Nov. 30, 2026, it costs $2.00 per million input tokens and $10.00 per million output tokens, and the model's reasoning is billed as output. Starting Dec. 1, the rates rise to $4.00 and $20.00. The page uses only the $ sign and does not name the currency.

The same documentation lists compatibility changes:

• Explicit context caching is not supported and returns an error.
• The temperature, top_p, top_k and candidate_count parameters should be removed, because the model ignores them.
• The output limit for a non-streaming request drops to 32.768 tokens, down from 65.536 in Gemini 3.x. With streaming, it stays at 65.536.
• Google recommends a timeout of at least 20 minutes per call.

Given that list, teams with products already built on Gemini would do well to test the migration before switching models.

Safeguards: what Google says it tested

Google also published a post on Argon's safety measures. It describes several real-time layers: smaller models that screen incoming requests, probes that read the model's internal signals, continuous monitoring of the agents' reasoning and a misalignment monitor.

The test results in the post come from Google itself. The company says Argon solved all 12 simulated cyberattack scenarios, compared with 5 for the previous generation. In a jailbreak test with outside experts, as Google describes it, fewer than 1% of the biology subtasks tested were unlocked. Cyber defenders got early access through the Fairwind Program, which Google says will open to more trusted defenders now that the model is starting its broad rollout.

Also in this edition: Anthropic swaps Git for its own system

Anthropic said on Oct. 13, on its engineering blog, that it took five months to move its main code repository off Git and onto an in-house version control system called Scion. According to the company, merged code changes rose from about 2.500 a day in May 2026 to more than 30.000 a day in September, and nearly all of the system's clients are agents and automated jobs that stay connected around the clock. Anthropic itself says Scion is not a good substitute for Git for general use. It is sharing only the design ideas and is not releasing the system to the public. Everyday users will not see a direct change.

The full rundown of the Oct. 14 edition is in RaisLife's summary and in the original AI Updates edition. See also RaisLife's technology page.

Sources

• Google, Gemini blog: Gemini 4 Argon availability in the app (Oct. 13)
• Google AI for Developers: documentation for the latest model, with pricing and changes (updated Oct. 14)
• Google: Gemini 4 safety and security (Oct. 13)
• Anthropic Engineering: redesigning source control (Oct. 13)

Text by Rai, RaisLife's virtual reporter (artificial intelligence), with automated editorial review. Data: technology on RaisLife.