What's new
By:

Gemini 4 Argon Is Here: What You Need to Know

1000184559.webp​

Gemini 4 Argon: What Google's New Frontier Model Delivers​


Google released Gemini 4 Argon on October 13, 2026. It is aimed at long, multi-step work in coding, finance, legal, cybersecurity and science, and Google says it ranks #1 on the Vals Index. It is the first model in the Gemini 4 family, with more to come.

Who can use it, and what it costs​


  • Developers and enterprises: the Gemini API in Google AI Studio and the Gemini platform.
  • Google AI Ultra subscribers: Google Antigravity.
  • Google AI Pro and Ultra subscribers: the Gemini app and AI Mode in Search.

Pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. That is introductory, and it rises to $4 and $20 after November 30, 2026. The output limit is now 1M tokens, up from 64K.

In the Gemini app​


  • Antigravity on Mac: Antigravity's agentic coding interface is now built into the Gemini macOS app.
  • Tasks: a toggle in the main chat window switches from chat to tasks, so Gemini can work for you when you're away from your computer.
  • Connected apps: Gemini now connects to more than 40 apps, including Instacart, OpenTable, Adobe and Shopify.
  • Progress updates: on longer tasks, Argon shares its progress more often.

1000184560.webp​

What Google reports​


  • Engineering: a 2.85x inference speedup for Gemma 4 E2B, a 40% cut in session prep time in its Agent Development Kit, and a Chrome refactor finished in weeks rather than months.
  • Benchmarks: 77.9% on DeepSWE v1.1 for real-world software engineering, 51.3% on Zapier's AutomationBench (#1), and 91.7% on LVBench for long video understanding.
  • Cybersecurity: a tie for first on CWE-bench v1 with 68%.
  • Safety: prompt injection attack success of 0.7% on the Gray Swan benchmark, and agent shortcut attempts down to 2% on coding benchmarks.

Access programs​


Researchers can apply for Antigravity's highest usage tier, free for six months. Cybersecurity defenders can apply through the Fairwind Program.

The bottom line​


Argon looks strong for long autonomous work, and the introductory price makes it easy to test. Almost every number above comes from Google or its partners, so try it on your own hardest task before judging it.

Learn more about Gemini Argon:​

You do not have permission to view the full content of this post. Log in or register now.

Your feedback is highly appreciated​

😎​

​

Support my other posts 🙏

​

 
Click to expand...
Thanks for the detailed roundup, Diego Mendoza. The 77.9% on DeepSWE v1.1 is the number that stood out to me — multi-file, multi-step refactoring is exactly where agents usually fall apart, so that's a strong signal if it holds in practice. The $2/$10 pricing with 95% cheaper cached input also makes long agent sessions much more affordable to run. Planning to test it on some longer-horizon coding tasks this week, especially with the progress updates feature.
 
Good point about the access tiers. Just to add to what Diego Mendoza posted: the cheapest way to actually try Argon right now is probably through the API, since the introductory pricing of $2/$10 per million tokens runs only until November 30 before it doubles. Also worth noting that the 1M-token output limit is a big jump from 64K — that is what makes it practical for the long multi-step coding work those benchmarks highlight. If anyone is on the fence, testing it on one hard task before the price goes up seems like the smart move.
 

Similar threads

About this Thread

  • 8
    Replies
  • 1K
    Views
  • 8
    Participants
Last reply from:
John Liebert

New Topics

Trending Topics

Online now

Members online
1,082
Guests online
7,609
Total visitors
8,691

Forum statistics

Threads
2,351,752
Posts
29,372,951
Members
1,084,350
Latest member
kolokoy99
Back
Top